no function in bmnist

* minor cleanups to improve import speed

* dumb

.github/workflows/test.yml

@@ -521,12 +521,11 @@ jobs:
       - name: Run HCQ2 tests
         run: HCQ_RUNTIME_DEV=PYTHON HCQ2=1 DEV=MOCKKFD+AMD FORWARD_ONLY=1 PYTHONPATH=. python test/test_tiny.py
       - name: Run HCQ2 multi-device tests
-        run: |
-          HCQ_RUNTIME_DEV=PYTHON HCQ2=1 DEV=MOCKKFD+AMD FORWARD_ONLY=1 PYTHONPATH=. python test/unit/test_multitensor.py \
-            TestMultiTensor.test_simple_add TestMultiTensor.test_shard_reduce \
-            TestMultiTensor.test_backward_sum TestMultiTensor.test_matmul_shard_0_0
+        run: HCQ_RUNTIME_DEV=PYTHON HCQ2=1 DEV=MOCKKFD+AMD FORWARD_ONLY=1 PYTHONPATH=. python -m pytest -n=auto test/backend/test_multitensor.py
       - name: Run HCQ2 JIT tests
         run: HCQ_RUNTIME_DEV=PYTHON HCQ2=1 DEV=MOCKKFD+AMD FORWARD_ONLY=1 PYTHONPATH=. python test/unit/test_jit.py
+      - name: Run HCQ2 unit tests
+        run: HCQ_RUNTIME_DEV=PYTHON HCQ2=1 DEV=MOCKKFD+AMD FORWARD_ONLY=1 PYTHONPATH=. python -m pytest test/device/test_hcq2.py
 
   testmockam:
     name: Linux (am)
@@ -589,7 +588,7 @@ jobs:
         if: ${{ matrix.backend == 'amd' && matrix.arch == 'gfx950' }}
         run: PYTHONPATH=. DEV=NULL:HIP:gfx950 MXFP4=1 LLAMA_LAYERS=2 BENCHMARK=3 NULL_ALLOW_COPYOUT=1 NO_HIPCC=1 ROCM_PATH=/opt/rocm JITBEAM=0 examples/mlperf/training_submission_v6.0/tinycorp/benchmarks/llama31_8b/implementations/tinybox_8xMI350X/profile.sh
       - name: Run pytest (amd)
-        run: python -m pytest -n=auto test/backend/test_ops.py test/backend/test_dtype.py test/backend/test_dtype_alu.py test/backend/test_linearizer.py test/backend/test_randomness.py test/backend/test_jit.py test/backend/test_graph.py test/backend/test_multitensor.py test/device/test_hcq.py test/external/external_test_am.py test/backend/test_asm_gemm.py::TestAsmGEMM test/opt/test_tensor_cores.py --durations=20
+        run: python -m pytest -n=auto test/backend/test_ops.py test/backend/test_dtype.py test/backend/test_dtype_alu.py test/backend/test_linearizer.py test/backend/test_randomness.py test/backend/test_jit.py test/backend/test_graph.py test/backend/test_multitensor.py test/device/test_hcq.py test/external/external_test_am.py test/backend/test_asm_gemm.py::TestAsmGEMM --durations=20
       - name: Run disk copy tests
         run: python -m pytest test/unit/test_disk_tensor.py -k test_copy_from_disk
       - name: Run TRANSCENDENTAL math

docs/kimi_k3_mi350.md

@@ -1,224 +0,0 @@
-# Kimi K3 on 8× MI350X
-
-This branch targets text generation directly from the official `moonshotai/Kimi-K3` checkpoint at `/raid/weights/kimi-k3`. It intentionally ignores the vision tower and multimodal projector. The checkpoint remains in its official 96-shard format; the loader never converts, rewrites, or creates a second 1.56 TB copy.
-
-The checked TP8 layout consumes 196.78 GB (183.27 GiB) of text weights per GPU. The compressed MLA cache adds 28.99 GB (27 GiB) per GPU at the full 1,048,576-token context, leaving approximately 62.23 GB of each nominal 288 GB MI350X for execution buffers and allocator overhead. Start much smaller.
-
-## Resume the current optimization session
-
-Work on branch `kimi_slop`. It was cleanly rebased onto `origin/kimi_slop` commit `553bdf68e` on 2026-08-10. The retained K3 commits after that base are `1b3732a6e`, `c6ac4961d`, `1d8620471`, `224bac031`, `f53f0e7e7`, and `2b1b8c22a`; verify the current hashes with `git log` because a later rebase may rewrite them. Before starting any benchmark, check that the worktree is clean and that no model process remains:
-
-```sh
-git status --short --branch
-git log --oneline --decorate -10
-pgrep -af 'tinygrad.llm.cli|benchmark_kimi_k3' || true
-```
-
-The active acceptance target is **more than 100 tok/s decode, more than 200 tok/s prefill, and less than 180 seconds cold startup** on TP8/gfx950. None is currently met. The authoritative official-checkpoint baseline is 389.84 seconds startup, 38.65 tok/s prefill, and 6.25 tok/s decode. The 1.56 TB checkpoint has a measured 6.9 GB/s single-XFS-NVMe read ceiling, giving a roughly 227-second physical cold-read floor; meeting the startup target therefore also requires a faster storage path, not only loader code.
-
-Use the fake-weight, one-layer loop for development. Do not repeatedly load the official checkpoint while optimizing:
-
-```sh
-DEV=AMD python extra/benchmark_kimi_k3_fake.py --mode attention --iterations 30
-DEV=AMD python extra/benchmark_kimi_k3_fake.py --mode block --iterations 30
-PROFILE=1 DEV=AMD python extra/benchmark_kimi_k3_fake.py --mode block --iterations 5
-```
-
-The clean retained baseline is about 0.630 ms per attention layer and 1.37 ms per complete block, with fake initialization taking about 0.9/2 seconds respectively after the rebase. Since K3 has 93 sequential blocks, a 100 tok/s projection requires at most approximately 0.108 ms per complete block. Only run another 96-shard official validation after a candidate produces a large whole-block gain, remains finite and deterministic, and passes a direct numerical comparison. Test one candidate at a time and remove failed experiments before moving on.
-
-The immediate bottleneck is launch and synchronization granularity: an official four-token decode profile contained 6,304 kernel events, while packed expert work was only a small fraction of total GPU time. Continue with whole-component or whole-block fusion/replay work, not isolated expert microkernels. The latest fake-loop A/B retested the previously rejected dual gate/up and weighted-down MFMA prototypes: 1.374 ms baseline versus 1.375 ms fused, so they were removed again. A fused whole-core KDA recurrence was also slower in the exact fake attention gate (0.665 versus 0.633 ms) and must not be restored unchanged.
-
-Preserve these invariants when official validation resumes: use `/raid/weights/kimi-k3` directly, keep all 96 shards byte-for-byte untouched, run only one model process, begin at context 128, verify all eight devices are `gfx950`, and preserve the first failure instead of retrying over it. The most recent preserved official failure from a rejected KDA experiment was the invalid sequence `[198, 163840, 163840, 163840]`; token 163840 is outside the valid vocabulary. The retained path before that experiment produced deterministic in-range replay.
-
-After a synthetic candidate passes, run correctness and performance in this order: NULL gfx950 compile coverage, focused tests with `-n12` where supported, TP8 fake numerical comparison, official context-128 deterministic tokens, load/prefill/decode timing, and then context admission at 4K, 32K, 131K, and 262K. Run `python -m mypy tinygrad/` and `python -m ruff check .` when those tools are installed. Read `tinygrad/viz/README.md` before inspecting rewrite or device profiles.
-
-## Before renting the machine
-
-- Keep the existing 96 shards in `/raid/weights/kimi-k3`; no additional model-sized free space is required. Leave ordinary headroom for logs and temporary files.
-- The host should have roughly 3 TB RAM, in line with AMD's MI350X platform guidance. The loader itself is streaming and must not need checkpoint-sized RAM.
-- Use a recent kernel/ROCm stack supported by the host vendor, although tinygrad uses its own AMD userspace driver when `DEV=AMD`.
-- Clone this exact commit/branch and keep the official checkpoint directory separate from the repository.
-
-Validate the existing directory without modifying it:
-
-```sh
-python examples/kimi_k3_prepare.py /raid/weights/kimi-k3 --context 4096
-```
-
-For a metadata-only preflight, place the official `config.json` and `model.safetensors.index.json` in a directory and run:
-
-```sh
-python examples/kimi_k3_prepare.py /raid/weights/kimi-k3 --metadata-only
-```
-
-## Hardware admission checks
-
-Do these before loading weights. Stop if any device is missing or reports a different architecture.
-
-```sh
-lspci -d 1002:75a0
-amd-smi list
-DEV=AMD DEBUG=2 python - <<'PY'
-from tinygrad import Device
-for i in range(8):
-  dev = Device[f"AMD:{i}"]
-  print(i, dev.arch)
-PY
-```
-
-Expected architecture: `gfx950` on all eight devices. Then run the small TP8 graph tests:
-
-```sh
-python -m pytest test/unit/test_llm_k3.py test/null/test_kimi_k3.py -q -n12
-DEV=NULL:HIP:gfx950 NULL_ALLOW_COPYOUT=1 python -m pytest \
-  test/unit/test_llm_k3.py::TestKimiK3::test_chunked_recurrent_generate -q -n1
-DEV=AMD python examples/kimi_k3_smoke.py --devices 8
-```
-
-The last two commands are deliberately small. They compile CDNA4 kernels and then exercise the complete TP8 topology without loading the checkpoint.
-
-For performance iteration, use the exact-width fake-weight harness before another official load:
-
-```sh
-DEV=AMD python extra/benchmark_kimi_k3_fake.py --mode attention --iterations 20
-DEV=AMD python extra/benchmark_kimi_k3_fake.py --mode block --iterations 20
-```
-
-It retains K3's 7,168-wide residual stream, 12,288-wide KDA state, 96 heads, 128×128 recurrent matrices, TP8 layouts, top-k 16 routing, packed MXFP4 expert shapes, collectives, and decode JIT, but uses one layer and 16 fake experts. Fake attention weights initialize in about 0.9 seconds and the full block in about 3 seconds. The retained path measured 0.630 ms per fake attention layer and 1.367 ms per complete fake block, projecting about 7.87 tok/s across 93 identical blocks versus 6.25 tok/s for the official heterogeneous model. Treat this as a candidate admission benchmark, not a correctness substitute for official weights.
-
-## First official load
-
-Start at a short context so cache allocation and compilation are bounded. The loader reads disk-backed safetensors, TP-shards every destination before realizing it, and drops each source shard/projection immediately afterward.
-
-```sh
-/usr/bin/time -v env DEV=AMD DEBUG=1 python -m tinygrad.llm.cli \
-  --model /raid/weights/kimi-k3 --devices 8 --max_context 128 </dev/null 2>&1 | tee kimi-k3-load.log
-```
-
-Watch host RAM, swap, HBM, temperatures, and XGMI traffic from a second terminal. Do not start with a one-million-token cache. If loading fails, preserve the first exception and the last loader progress line; do not retry with a larger host-side cache.
-
-## Correctness and performance sequence
-
-1. Load with context 128 and generate one token.
-2. Repeat a fixed prompt twice and confirm token-for-token deterministic greedy output.
-3. Compare the first several greedy tokens against the official Transformers implementation at temperature zero.
-4. Benchmark decode only after two warm-up tokens.
-5. Benchmark prefill at 128, 512, 2K, and 8K tokens. Increase context only while HBM and compile time remain healthy.
-6. Use `VIZ=1` plus `python -m tinygrad.viz.cli` to inspect kernels; use `VIZ=2` only for short SQTT captures because it adds overhead.
-
-Example decode benchmark:
-
-```sh
-DEV=AMD DEBUG=1 python -m tinygrad.llm.cli --model /raid/weights/kimi-k3 \
-  --devices 8 --max_context 4096 --warmup --benchmark 20
-```
-
-## MI350X validation results (2026-08-10)
-
-The official directory was audited in place: 96 shards, 497,220 indexed tensors, 497,052 language tensors, and 1,560,860,324,864 total bytes. All eight devices reported `gfx950`. No checkpoint file was converted, copied, or modified, and every model run used a single process. The actual text tower is 1,559,965,606,912 bytes; its checked TP8 layout is 196,784,397,312 bytes per GPU.
-
-The preserved first full-checkpoint error was an `A_log` shape mismatch, `(128,) -> (96, 1)`. K3 stores one decay value per 128-wide KDA channel, not one per head. The loader now keeps this field replicated and applies the official channel-wise broadcast. A numerical unit test covers the distinction from the older head-wise Kimi Linear behavior.
-
-Load speed was fixed before generation. The original loader opened thousands of individual expert tensors and independently realized eight strided TP slices. The MI350 path now does the following without changing the checkpoint:
-
-- parses safetensor headers selectively, constructing disk-backed tensors only for the 2,460 non-expert entries consumed by that pass instead of materializing metadata objects for every expert entry twice;
-- copies contiguous axis-zero shards and replicas directly into their final device buffers;
-- reads a replicated tensor once and fans it out over XGMI instead of issuing eight identical direct reads (14.31 GB less RAID traffic);
-- stages an inner-axis tensor once and schedules all eight TP slices together;
-- reads each layer's contiguous 15.72 GB expert region once, reorders its lexicographically stored expert records on GPU 0, and realizes all six packed/scale destinations together;
-- retains only final MultiBuffer identities, drops the reorder graph, and flushes the 15.72 GB staging allocation before the next layer.
-
-One real expert layer leaves exactly 1,965,293,568 bytes resident on each GPU and zero bytes in the GPU-0 allocator cache. Complete context-128 loads measured 527.20 seconds before the final staging cleanup and 490.05/489.59 seconds afterward. Peak host RSS for the unprofiled correctness run was 2.11 GiB with zero swap. RAID variability produced later loads from 489.06 to 532.85 seconds.
-
-The selective-metadata and bounded-GC pass reduced non-expert loading from 125.77 to 57.77 seconds. A subsequent full official context-128 load completed in 411.49 seconds, 78.10 seconds (16.0%) faster than the 489.59-second baseline. It read the 96 shards in place with 1,049,688 KiB peak host RSS and zero swap; no weight payload was converted, copied, or modified. Direct-I/O probes measured approximately 6.9 GB/s aggregate for both one and eight concurrent 1 GiB reads. At that rate the 1.56 TB checkpoint has a roughly 227-second cold-read lower bound, so this RAID cannot meet a true cold sub-three-minute startup regardless of loader overhead.
-
-Expert staging graphs are acyclic and are released by reference counting after each layer, so the loader now suppresses unnecessary cyclic-collector scans only around that loop and restores its prior state on every exit. A quiet context-128 load then completed in 391.54 seconds, 30.14 seconds (7.1%) faster than the immediately preceding 421.68-second run, with 1.04 GiB peak RSS and zero swap, although storage variability contributes to run-to-run timing. The host used for these measurements actually mounts `/raid` from one 3.5 TB XFS NVMe, not a multi-drive RAID; shard 28 has 218 extents and live reads fell to roughly 160 MB/s there. This storage layout, plus the physical checkpoint size, remains the limiting cold-start constraint. The weights were not defragmented, copied, or modified.
-
-The fixed XTML prompt `Reply with exactly: OK` encodes to 93 tokens. After excluding the cold JIT capture from replay comparison, two greedy runs produced the identical eight-token sequence:
-
-```text
-[9545, 59991, 10580, 14404, 9545, 59991, 9545, 59991]
-```
-
-At context 128, steady prefill was 14.32 seconds (6.49 tok/s) and eight-token decode was 2.27 seconds (3.53 tok/s, 283.3 ms/token). The same first tokens remained stable at every admitted context. These rates are much lower than the planning estimates below and should be treated as the current measured baseline.
-
-The retained gfx950 serving pass enables the validated wave64 recurrent prefill kernel with 128-token chunks, uses exact BF16 decode projections, combines the routed/shared final TP partials into one collective, and tiles four adjacent packed-expert outputs during multi-token execution. On the same 93-token prompt, two replay trials produced the identical sequence `[198, 92652, 220, 80225]`. Prefill replay measured 2.418--2.482 seconds (37.47--38.46 tok/s), and eight-token decode measured 1.294 seconds (6.18 tok/s, 161.81 ms/token). Peak RSS was 2.77 GiB with zero swap. The packed prefill tile changes floating-point reduction order: direct official-layer comparison against the original kernel had maximum differences of 0.015625 for gate and 0.0078125 for down, and the end-to-end greedy sequence was stable across replay.
-
-A subsequent gfx950 decode pass split the 7,168-wide replicated BF16 projections across eight waves per 16 output channels and used CDNA4 BF16 MFMA, with one FP32 LDS reduction at the end. It is enabled only for batch-one/token-one replicated projections whose dimensions satisfy the hardware tile; prefill, the FP32 router, and the output-sharded 12,288-wide KDA gate remain unchanged. The official retained path uses it for MLA q-a/kv-a and KDA f-a. Isolated TP8 measurements improved replicated 128/576-output projections by about 16--18%; applying it to the already output-sharded KDA gate was slower and was rejected. Random-shape comparison against the generic graph had maximum/mean absolute BF16 differences of 2.0/0.1114 because the split changes reduction order. Against a serial FP32 accumulation rounded once to BF16, the 7,168-to-1,536 kernel was bit-exact in the tested sample.
-
-The final official context-128 validation loaded in 389.84 seconds with 2.71 GiB peak RSS and zero swap. Two replay trials produced the identical four-token sequence `[198, 59675, 9817, 12519]`; prefill remained 2.406 seconds (38.65 tok/s), while eight-token decode improved to 1.280 seconds (6.25 tok/s, 160.00 ms/token). A one-wave MFMA variant and a full-wave fused decode recurrence were both rejected: the former delivered 6.02 tok/s, and the latter 6.179 tok/s, while both changed the greedy sequence without a useful speed gain.
-
-A final load-first experiment increased the disk-to-HBM io_uring queue depth from one to the 32 existing bounded 2 MiB staging buffers. On a direct 1 GiB read from fragmented shard 28 it measured 6.834 GB/s versus 6.832 GB/s for the original path, so the change was rejected. The subsequent unmodified official 96-shard load completed in 389.48 seconds, confirming both the prior result and the single-NVMe lower bound. Peak RSS was 2.75 GiB with zero swap.
-
-Two direct packed-expert MFMA prototypes were also rejected after that load. A fused gate/up kernel was about 29% faster in isolation at the TP8-local shape, and a routed-down kernel which combined projection, probability weighting, and route reduction measured 1.45 ms versus 2.42 ms in isolation. End-to-end, however, stable replay produced `[198, 2338, 2127, 148297]`, prefill measured 38.87 tok/s, and decode measured 6.263 tok/s. That is indistinguishable from the retained 38.65/6.25 tok/s path while changing floating-point reduction order, so neither kernel was retained.
-
-A whole-core KDA decode experiment fused convolution, Q/K normalization, channel decay, recurrence, RMS normalization, output gating, and four persistent state updates. Its raw kernel replayed in about 109 microseconds per local KDA layer and matched a one-step synthetic reference within `9.77e-4` output and `8.13e-4` state maximum error. The exact-width fake-layer gate caught that it was slower than the retained attention path (0.665 versus 0.633 ms/layer). The already-running official validation was stopped after its first invalid greedy sequence, `[198, 163840, 163840, 163840]`, where 163840 is outside the checkpoint's vocabulary. The kernel was rejected and removed.
-
-| Maximum context | Load | Short-prompt replay | Result |
-|---:|---:|---:|---|
-| 128 | 489.59s | 14.32s | stable 8-token replay |
-| 4,096 | 489.06s | 14.32s | stable replay, zero swap |
-| 32,768 | 532.85s | 14.33s | stable replay, zero swap |
-| 131,072 | 520.91s | 14.37s | stable first token, zero swap |
-| 262,144 | 497.34s | 14.41s | stable first token, zero swap |
-
-These are maximum-context/cache admission tests with the same 93-token prompt, not full-length 32K/131K/262K prefills. The full cache allocation path was exercised, but filling those contexts remains a separate long-running throughput test.
-
-Runtime profiling bracketed four steady decode tokens. It recorded 6,304 kernel events and about 474--478 ms of summed GPU work across the eight devices inside a roughly 1.5-second profiled wall interval. The packed `mxfp4_expert_linear_wave64` kernels accounted for only about 22.5 ms summed; the largest families were small 1,792-wide reductions. This identifies launch/synchronization granularity as the immediate MI350 bottleneck rather than packed-weight bandwidth. `JIT_BATCH_SIZE=64` produced the same original 3.53 tok/s as 32. A gfx950 fused MXFP8 QDQ experiment was bit-exact but slower on the real device (about 95 microseconds versus 57--64 microseconds), so it was rejected. Combining the routed and shared final TP partials removed one collective per routed decode layer and helped raise unprofiled decode to 6.18 tok/s, but the remaining sequential launch boundaries still dominate.
-
-The checkpoint's bundled Transformers code was used as the architectural reference for channel decay and tensor mapping. A full independent Transformers/vLLM token comparison was not run on this host because the required `compressed_tensors`/serving backend is not installed; deterministic tinygrad replay and the numerical KDA, loader-layout, NULL gfx950 compile, and real TP8 smoke tests are the completed correctness gates.
-
-## Known hardware-only gate
-
-The correctness path now consumes packed MXFP4 expert weights directly on gfx950 with a wave64 software-decode kernel, so it does not create selected-expert BF16 weight expansions. MXFP8 activation quantization is still emulated. tinygrad has gfx950/CDNA4 BF16 and FP8 matrix-core support, but this branch does not yet have a hardware-validated native MXFP4×MXFP8 expert GEMM. Expect the first run to be a correctness bring-up, not production throughput. Capture profiles on MI350X before changing the representation: native FP4 work cannot be validated faithfully on the available gfx1100 cards.
-
-Recurrent prefill is fused. The gfx950 wave-parallel kernel was compared directly with the portable graph at the official per-GPU shape through 128 tokens: maximum core/state differences remained below `8e-6`/`1e-6`, outputs were finite, and replay was about 2.7 ms versus about 8 ms for the portable kernel in the isolated test. Full K3 therefore uses 128-token recurrent chunks on gfx950. Chunk size remains part of the numerical configuration because different reduction orders can select different final greedy tokens.
-
-The following serving changes apply to the official K3 path: recurrent-state reset graph capture, direct AMD scalar readback without rebuilding a scheduler graph, materialized gate/up boundaries, separate greedy decode JITs, K3's uncorrected routed probability semantics, gfx950 KDA Q/K/V and exact BF16 partial projections, one combined routed/shared final collective, a gfx950 greedy output-head kernel, the wave64 packed-expert path, and the multi-token four-output packed tile. Software MXFP8 remains in use.
-
-After hardware admission on MI350X, profile before porting those kernels. The likely implementation order is:
-
-1. A native packed MXFP4×MXFP8 grouped expert GEMM using CDNA4 matrix instructions.
-2. A wave64/MFMA KDA Q/K/V decode projection.
-3. Combined routed/shared down-projection TP partials so each layer performs one XGMI all-reduce.
-4. A CDNA4 output-head matvec and router matvec if they remain visible in the profile.
-
-Every port needs a direct numerical comparison with the generic graph and an end-to-end greedy-token comparison before performance measurements. The wave64 packed-expert kernel has compile coverage through `NULL:HIP:gfx950`; numerical and performance validation still require real MI350X hardware. None of the remaining gfx11-only kernels should be enabled on gfx950 by changing only the architecture guard.
-
-## MI350X performance expectation
-
-Treat the first rental as bring-up, not a guaranteed throughput run. The loader reads every official expert tensor once into a transient GPU-0 staging buffer (at most one packed projection), then redistributes TP8 slices over the GPU fabric; it does not generate files or require checkpoint-sized host RAM. A reasonable planning range for the full text model on eight MI350X cards is 3–8 minutes to stream and TP-shard the 1.56 TB checkpoint, 150–400 tok/s for initial short/medium prefill, and 25–60 tok/s decode with the software packed-expert path. After a native CDNA4 MXFP4×MXFP8 grouped expert kernel, wave64/MFMA recurrent projections, and XGMI collective tuning, 500+ tok/s prefill and roughly 80–150 tok/s decode are plausible targets. These ranges are engineering estimates, not measurements.
-
-The nominal HBM bandwidth is not the main uncertainty: eight MI350X devices have enough aggregate bandwidth for K3's active weights. Utilization is limited by 93 sequential layers, small routed projections, and synchronization after TP input-sharded projections. Record actual HBM and XGMI counters before deciding whether the next port should target matrix instructions or collective count.
-
-The official checkpoint also contains MoonViT-V2 and multimodal projector weights. They are skipped by the text loader. Image input remains a separate implementation and validation task.
-
-## Local TP4 performance baseline
-
-The pre-rental benchmark uses the converted `Kimi-Linear-48B-A3B-Instruct-MXFP4-v2` checkpoint on four gfx1100 GPUs. It is a useful regression test for the KDA/MLA/MoE text path, not a projection of K3 throughput on MI350X.
-
-```sh
-DEV=AMD JIT_BATCH_SIZE=64 python extra/benchmark_kimi.py \
-  /raid/models/Kimi-Linear-48B-A3B-Instruct-MXFP4-v2 \
-  --devices 4 --max-context 128 --prompt-tokens 32 --decode-tokens 32 --chunk-size 32
-```
-
-Results from 2026-08-10:
-
-- load from RAID: 44.28s for the 29.27 GB checkpoint
-- first 32-token prefill includes roughly 10s of compilation/capture
-- steady fresh-prompt prefill replay: 0.118s, 270.20 tok/s
-- steady context-32 decode replay: 101.82 tok/s, 9.82 ms/token
-- peak host RSS: 729.9 MiB; swap was not used
-
-The load, prefill, and decode targets are all met in the bounded prompt-32 run. Decode improved from 23.03 tok/s to 101.82 tok/s. The retained greedy output was checked across 32 decode steps; rejected half-wave and unrounded recurrent reductions were faster but diverged and eventually collapsed to a repeated token.
-
-Fully warmed HTTP serving was also measured with `--max_context 4096`. Startup, including weight load, capture, and replay of both serving shapes, took 113.73s. After a two-turn cache test, the first aligned 64-token request reported 271 tok/s prefill and 101 tok/s decode over 64 generated tokens. A 99-token prompt reported 254 tok/s prefill and 99 tok/s decode; decode falls slightly as MLA context grows.
-
-Recurrent serving uses only the captured 32-token prefill graph and captured single-token graph. Warmup uses two consecutive chunks so both initial and nonzero-position prefill execution are ready before the socket opens. A prompt tail shorter than 32 tokens runs through the single-token graph instead of compiling a new static shape, so no request-time JIT capture is required. Exact extensions reuse recurrent and KV state—the live second turn logged `in: 18 + 15`—while divergent prompts reset both safely. Very short prompts can report less than 200 aggregate prefill tok/s because fixed reset and single-token costs dominate; aligned and medium/long prompts exercise the 200+ tok/s prefill path.
-
-Four 7900 XTX cards provide 96 GB aggregate VRAM and about 3.84 TB/s aggregate physical memory bandwidth. Their nominal aggregate vector FP16 rate is about 245.6 TFLOP/s, or about 492 TFLOP/s through matrix instructions. Kimi Linear activates roughly 3.107B parameters per token; a simple active-weight accounting gives approximately 4.05 GB/token and an optimistic bandwidth-only ceiling near 948 tok/s. The measured decode rate is much lower because this MoE decode workload is a collection of small matrix-vector operations plus PCIe collectives, not one ideal streaming kernel.
-
-The generic loader currently rereads logical TP shards and accounts for roughly 227 GB of disk traffic for a TP4 load. RAID bandwidth hides that inefficiency locally, but a direct one-pass shard loader remains worthwhile before slow remote storage is used. It was not retained here because the attempted direct-shard graph exposed an unresolved scheduler/renderer edge; correctness and bounded memory take priority over avoiding the redundant reads.
-
-Different chunk sizes can choose a different final token because their matrix kernels use different floating-point reduction orders. Each measured shape was repeatable between cold and captured execution. For official K3 validation, compare logits/tokens against the reference at one fixed chunk size and greedy settings rather than requiring bitwise agreement between performance shapes.

examples/beautiful_mnist.py

@@ -1,6 +1,6 @@
 # model based off https://medium.com/data-science/going-beyond-99-mnist-handwritten-digits-recognition-cfff96337392
 from typing import Callable
-from tinygrad import Tensor, TinyJit, nn, GlobalCounters, function, Context
+from tinygrad import Tensor, TinyJit, nn, GlobalCounters, Context
 from tinygrad.helpers import getenv, colored, trange
 from tinygrad.nn.datasets import mnist
 
@@ -15,7 +15,6 @@ class Model:
       nn.BatchNorm(64), Tensor.max_pool2d,
       lambda x: x.flatten(1), nn.Linear(576, 10)]
 
-  @function
   def __call__(self, x:Tensor) -> Tensor: return x.sequential(self.layers)
 
   @TinyJit

examples/convert_kimi_linear.py

@@ -1,9 +0,0 @@
-import argparse
-from tinygrad.llm.kimi import convert_kimi
-
-if __name__ == "__main__":
-  parser = argparse.ArgumentParser(description="Convert official Kimi-Linear-48B-A3B BF16 weights to tinygrad MXFP4/BF16")
-  parser.add_argument("source", help="downloaded moonshotai/Kimi-Linear-48B-A3B-Instruct directory")
-  parser.add_argument("output", help="output directory")
-  args = parser.parse_args()
-  convert_kimi(args.source, args.output)

examples/kimi_k3_prepare.py

@@ -1,27 +0,0 @@
-#!/usr/bin/env python3
-"""Cheap preflight for an official moonshotai/Kimi-K3 checkout. Does not load model weights."""
-import argparse, json, pathlib, shutil
-from tinygrad.llm.kimi_k3 import KIMI_K3_TP8_BYTES_PER_GPU, audit_kimi_k3_checkpoint
-
-def main() -> None:
-  parser = argparse.ArgumentParser()
-  parser.add_argument("model_dir", type=pathlib.Path)
-  parser.add_argument("--metadata-only", action="store_true", help="permit absent weight shards")
-  parser.add_argument("--context", type=int, default=4096, help="context length used for the memory estimate")
-  args = parser.parse_args()
-  stats = audit_kimi_k3_checkpoint(args.model_dir, require_shards=not args.metadata_only)
-  if not 1 <= args.context <= 1_048_576: raise ValueError("--context must be between 1 and 1048576")
-
-  # K3 has 24 MLA layers. Each token stores the 512-value compressed latent plus 64 RoPE values in BF16.
-  per_gpu_weights = KIMI_K3_TP8_BYTES_PER_GPU
-  mla_cache = 24 * args.context * (512 + 64) * 2
-  hbm = 288_000_000_000
-  print(json.dumps(stats, indent=2))
-  print(f"exact text weights/GPU under this TP8 layout: {per_gpu_weights/1e9:.2f} GB ({per_gpu_weights/2**30:.2f} GiB)")
-  print(f"replicated MLA cache/GPU at {args.context:,} tokens: {mla_cache/1e9:.2f} GB ({mla_cache/2**30:.2f} GiB)")
-  print(f"nominal MI350X headroom before runtime buffers: {(hbm-per_gpu_weights-mla_cache)/1e9:.2f} GB")
-  if not args.metadata_only:
-    usage = shutil.disk_usage(args.model_dir)
-    print(f"filesystem free space: {usage.free/1e9:.2f} GB")
-
-if __name__ == "__main__": main()

examples/kimi_k3_smoke.py

@@ -1,27 +0,0 @@
-#!/usr/bin/env python3
-"""Run a reduced, architecture-complete K3 prefill/decode on tensor-parallel devices."""
-import argparse, time
-from tinygrad import Tensor, Device, dtypes, nn
-from tinygrad.llm.kimi_k3 import _shard_kimi_k3, kimi_k3_smoke_config
-from tinygrad.llm.model import Transformer
-
-def main() -> None:
-  parser = argparse.ArgumentParser()
-  parser.add_argument("--devices", type=int, default=8)
-  args = parser.parse_args()
-  if args.devices not in (1, 2, 4, 8): raise ValueError("the K3 admission smoke test supports 1, 2, 4, or 8 devices")
-  devices = tuple(f"AMD:{i}" for i in range(args.devices))
-  model = Transformer(kimi_k3_smoke_config())
-  for name,value in nn.state.get_state_dict(model).items():
-    fill = 127 if name.endswith("weight_scale") else 0
-    dtype = value.dtype if value.dtype is dtypes.uint8 else dtypes.bfloat16
-    value.replace(Tensor.full(value.shape, fill, dtype=dtype, device="CPU"))
-  _shard_kimi_k3(model, devices)
-  temperature = Tensor([0.0], device=devices)
-  for label,tokens,start in (("prefill", [[1, 2]], 0), ("decode", [[3]], 2), ("decode replay", [[4]], 3)):
-    begin = time.perf_counter()
-    out = model(Tensor(tokens, dtype=dtypes.int32, device=devices), start, temperature).realize()
-    for device in devices: Device[device].synchronize()
-    print(f"{label}: shape={out.shape}, {time.perf_counter()-begin:.3f}s")
-
-if __name__ == "__main__": main()

examples/mlperf/training_submission_v6.0/tinycorp/benchmarks/llama31_8b/implementations/tinybox_8xMI350X/dev_beam.sh

@@ -26,7 +26,7 @@ export FUSED_SILU_W13=${FUSED_SILU_W13:-1}
 export SPLIT_W13=${SPLIT_W13:-0}
 export OFFLOAD_OPTIM=${OFFLOAD_OPTIM:-0}
 
-export DEFAULT_FLOAT="bfloat16" OPTIM_DTYPE="bfloat16"
+export DEFAULT_FLOAT="bfloat16" OPTIM_DTYPE="float32"
 export DP=${DP:-8} MP=${MP:-1} BS=${BS:-16} EVAL_BS=${EVAL_BS:-8} GRADIENT_ACC_STEPS=${GRADIENT_ACC_STEPS:-2}
 export GBS=$((BS * GRADIENT_ACC_STEPS))
 

examples/mlperf/training_submission_v6.0/tinycorp/benchmarks/llama31_8b/implementations/tinybox_8xMI350X/dev_run.sh

@@ -26,7 +26,7 @@ export FUSED_SILU_W13=${FUSED_SILU_W13:-1}
 export SPLIT_W13=${SPLIT_W13:-0}
 export OFFLOAD_OPTIM=${OFFLOAD_OPTIM:-0}
 
-export DEFAULT_FLOAT="bfloat16" OPTIM_DTYPE="bfloat16"
+export DEFAULT_FLOAT="bfloat16" OPTIM_DTYPE="float32"
 export DP=${DP:-8} MP=${MP:-1} BS=${BS:-16} EVAL_BS=${EVAL_BS:-8} GRADIENT_ACC_STEPS=${GRADIENT_ACC_STEPS:-2}
 export GBS=$((BS * GRADIENT_ACC_STEPS))
 

extra/benchmark_kimi.py

@@ -1,79 +0,0 @@
-#!/usr/bin/env python3
-"""Benchmark Kimi-Linear load, prefill, and decode on its TP4 checkpoint."""
-import argparse, resource, time
-from tinygrad import Device, TinyJit
-from tinygrad.helpers import profile_marker
-from tinygrad.llm.kimi import load_kimi
-
-def sync(devices:int) -> None:
-  for i in range(devices): Device[f"AMD:{i}"].synchronize()
-
-def timed_next(gen, devices:int) -> tuple[int, float]:
-  begin = time.perf_counter()
-  token = next(gen)
-  sync(devices)
-  return token, time.perf_counter()-begin
-
-def fresh_generate(model, prompt:list[int], chunk_size:int):
-  # Force recurrent/KV state reset so repeated runs and chunk sweeps measure the entire prompt,
-  # rather than silently reusing the prefix cached by the previous measurement.
-  model._cached_tokens = [-1] * len(prompt)
-  return model.generate(prompt.copy(), chunk_size=chunk_size)
-
-def main() -> None:
-  parser = argparse.ArgumentParser()
-  parser.add_argument("model", help="converted Kimi-Linear-48B-A3B MXFP4-v2 directory")
-  parser.add_argument("--devices", type=int, default=4)
-  parser.add_argument("--max-context", type=int, default=128)
-  parser.add_argument("--prompt-tokens", type=int, default=32)
-  parser.add_argument("--decode-tokens", type=int, default=8)
-  parser.add_argument("--chunk-size", type=int, default=32)
-  parser.add_argument("--sweep-chunks", help="comma-separated prefill chunk sizes; uses the fastest for decode")
-  args = parser.parse_args()
-  if args.prompt_tokens < 1 or args.prompt_tokens + args.decode_tokens + 1 > args.max_context:
-    raise ValueError("prompt and decode tokens must fit within --max-context")
-
-  begin = time.perf_counter()
-  model = load_kimi(args.model, max_context=args.max_context, devices=args.devices)
-  sync(args.devices)
-  print(f"load: {time.perf_counter()-begin:.3f}s", flush=True)
-
-  prompt = [1] + [1000+i%1000 for i in range(args.prompt_tokens-1)]
-  chunks = [int(x) for x in args.sweep_chunks.split(",")] if args.sweep_chunks else [args.chunk_size]
-  if any(x < 1 or x > args.prompt_tokens for x in chunks): raise ValueError("prefill chunks must be between 1 and --prompt-tokens")
-  timings:list[tuple[float, int]] = []
-  prefill_jits:dict[int, TinyJit] = {}
-  for chunk in chunks:
-    # Recurrent prefill has a static token dimension. Give each swept shape its own capture;
-    # the rollout JIT remains shared and independently benchmarks chunk 1/decode.
-    if chunk != 1: model.prefill_jit = TinyJit(model.forward)
-    cold = fresh_generate(model, prompt, chunk)
-    first, cold_prefill = timed_next(cold, args.devices)
-    print(f"chunk {chunk}: cold prefill {cold_prefill:.3f}s, token={first}", flush=True)
-    warm = fresh_generate(model, prompt, chunk)
-    warm_first, prefill = timed_next(warm, args.devices)
-    if first != warm_first: raise RuntimeError(f"chunk {chunk} is not repeatable: cold={first}, warm={warm_first}")
-    timings.append((prefill, chunk))
-    if chunk != 1: prefill_jits[chunk] = model.prefill_jit
-    print(f"chunk {chunk}: prefill {prefill:.3f}s ({args.prompt_tokens/prefill:.3f} tok/s), token={first}", flush=True)
-
-  prefill, best_chunk = min(timings)
-  if best_chunk != 1: model.prefill_jit = prefill_jits[best_chunk]
-  warm = fresh_generate(model, prompt, best_chunk)
-  first, replay_prefill = timed_next(warm, args.devices)
-  _, cold_decode = timed_next(warm, args.devices)
-  _, capture_decode = timed_next(warm, args.devices)
-  print(f"selected chunk: {best_chunk}; prefill replay {replay_prefill:.3f}s "
-        f"({args.prompt_tokens/replay_prefill:.3f} tok/s), token={first}", flush=True)
-  print(f"cold decode: {cold_decode:.3f}s", flush=True)
-  print(f"capture decode: {capture_decode:.3f}s", flush=True)
-  profile_marker("kimi decode steady start")
-  begin = time.perf_counter()
-  output = [next(warm) for _ in range(args.decode_tokens)]
-  sync(args.devices)
-  decode = time.perf_counter()-begin
-  profile_marker("kimi decode steady end")
-  print(f"decode: {decode:.3f}s ({args.decode_tokens/decode:.3f} tok/s, {decode/args.decode_tokens*1e3:.3f} ms/tok), output={output}", flush=True)
-  print(f"peak RSS: {resource.getrusage(resource.RUSAGE_SELF).ru_maxrss/1024:.1f} MiB", flush=True)
-
-if __name__ == "__main__": main()

extra/benchmark_kimi_k3.py

@@ -1,83 +0,0 @@
-#!/usr/bin/env python3
-"""Bounded correctness and load/prefill/decode benchmark for the official TP8 Kimi K3 checkpoint."""
-import argparse, resource, time
-
-from tinygrad import Device
-from tinygrad.helpers import profile_marker
-from tinygrad.llm.cli import KimiK3Template, SimpleTokenizer
-from tinygrad.llm.kimi_k3 import load_kimi_k3, load_kimi_tokenizer_data
-
-def sync(devices:int) -> None:
-  for i in range(devices): Device[f"AMD:{i}"].synchronize()
-
-def fresh_generate(model, prompt:list[int], chunk_size:int):
-  # Never reuse a prefix or recurrent state across correctness/benchmark trials.
-  model._cached_tokens = [-1] * len(prompt)
-  return model.generate(prompt.copy(), chunk_size=chunk_size, temperature=0.0)
-
-def timed_next(gen, devices:int) -> tuple[int, float]:
-  begin = time.perf_counter()
-  token = next(gen)
-  sync(devices)
-  return token, time.perf_counter()-begin
-
-def main() -> None:
-  parser = argparse.ArgumentParser()
-  parser.add_argument("model", help="official unmodified Kimi K3 checkpoint directory")
-  parser.add_argument("--devices", type=int, default=8)
-  parser.add_argument("--max-context", type=int, default=128)
-  parser.add_argument("--prompt", default="Reply with exactly: OK")
-  parser.add_argument("--stable-tokens", type=int, default=8)
-  parser.add_argument("--decode-tokens", type=int, default=8)
-  parser.add_argument("--chunk-size", type=int, default=128)
-  args = parser.parse_args()
-
-  begin = time.perf_counter()
-  model = load_kimi_k3(args.model, max_context=args.max_context, devices=args.devices)
-  sync(args.devices)
-  load_time = time.perf_counter()-begin
-  print(f"load: {load_time:.3f}s", flush=True)
-
-  normal, special, bos, eos = load_kimi_tokenizer_data(args.model)
-  tok = SimpleTokenizer(normal, special, "kimi-k2", bos_id=bos, eos_id=eos, eot_id=eos)
-  rendered = KimiK3Template().render(messages=[{"role":"user", "content":args.prompt}], add_generation_prompt=True)
-  prompt = tok.encode(rendered)
-  needed = len(prompt) + max(args.stable_tokens, args.decode_tokens+3)
-  if needed > args.max_context: raise ValueError(f"prompt and output need {needed} tokens but max context is {args.max_context}")
-  print(f"prompt: {len(prompt)} tokens, chunk={args.chunk_size}", flush=True)
-
-  sequences:list[list[int]] = []
-  # TinyJit executes uncaptured once, captures the second call, and replays from the third call.
-  # Compare two replay paths rather than capture numerics/timing against replay.
-  for trial in range(4):
-    gen = fresh_generate(model, prompt, args.chunk_size)
-    sequence:list[int] = []
-    prefill = 0.0
-    for step in range(args.stable_tokens):
-      token, elapsed = timed_next(gen, args.devices)
-      sequence.append(token)
-      if step == 0: prefill = elapsed
-    if trial >= 2: sequences.append(sequence)
-    label = ("uncaptured warmup", "capture warmup", "stable trial 1", "stable trial 2")[trial]
-    print(f"{label}: prefill={prefill:.3f}s "
-          f"({len(prompt)/prefill:.3f} tok/s), tokens={sequence}", flush=True)
-  if sequences[0] != sequences[1]: raise RuntimeError(f"greedy output is not repeatable: {sequences}")
-  print(f"stable text: {tok.decode(sequences[0])!r}", flush=True)
-
-  gen = fresh_generate(model, prompt, args.chunk_size)
-  profile_marker("kimi k3 steady prefill start")
-  first, prefill = timed_next(gen, args.devices)
-  profile_marker("kimi k3 steady prefill end")
-  warmup = [timed_next(gen, args.devices)[0] for _ in range(2)]
-  profile_marker("kimi k3 steady decode start")
-  begin = time.perf_counter()
-  output = [next(gen) for _ in range(args.decode_tokens)]
-  sync(args.devices)
-  decode = time.perf_counter()-begin
-  profile_marker("kimi k3 steady decode end")
-  print(f"prefill replay: {prefill:.3f}s ({len(prompt)/prefill:.3f} tok/s), token={first}", flush=True)
-  print(f"decode after warmup {warmup}: {decode:.3f}s ({args.decode_tokens/decode:.3f} tok/s, "
-        f"{decode/args.decode_tokens*1e3:.3f} ms/tok), output={output}", flush=True)
-  print(f"peak RSS: {resource.getrusage(resource.RUSAGE_SELF).ru_maxrss/1024:.1f} MiB", flush=True)
-
-if __name__ == "__main__": main()

extra/benchmark_kimi_k3_fake.py

@@ -1,84 +0,0 @@
-#!/usr/bin/env python3
-"""Fast exact-shape K3 KDA/layer benchmark using bounded fake weights instead of the 1.56 TB checkpoint."""
-from __future__ import annotations
-import argparse, statistics, time
-from dataclasses import replace
-
-from tinygrad import Device, Tensor, TinyJit, dtypes, nn
-from tinygrad.helpers import profile_marker
-from tinygrad.llm.kimi_k3 import kimi_k3_config
-from tinygrad.llm.model import GatedDeltaNetBlock
-
-def tp_axis(name:str) -> int|None:
-  if "ffn_gate_exps.weight" in name or "ffn_up_exps.weight" in name: return 1
-  if "ffn_gate_exps.weight_scale" in name or "ffn_up_exps.weight_scale" in name: return 1
-  if "ffn_down_exps.weight" in name or "ffn_down_exps.weight_scale" in name: return 2
-  if name.endswith(("ffn_gate_shexp.weight", "ffn_up_shexp.weight")): return 0
-  if name.endswith(("ffn_down_shexp.weight", "ffn_routed_down.weight", "ffn_routed_up.weight", "ssm_out.weight")): return 1
-  if name.endswith(("attn_q.weight", "attn_k.weight", "attn_v.weight", "ssm_g_full.weight", "ssm_f_b.weight", "ssm_beta.weight")): return 0
-  if name.endswith(("ssm_q_conv1d.weight", "ssm_k_conv1d.weight", "ssm_v_conv1d.weight", "ssm_dt.bias")): return 0
-  return None
-
-def fake_value(name:str) -> tuple[int|float, object]:
-  if name.endswith("weight_scale"): return 120, dtypes.uint8
-  if name.endswith("_exps.weight"): return 0x11, dtypes.uint8
-  if name.endswith("ssm_a"): return -0.1, dtypes.float32
-  if name.endswith("ssm_dt.bias"): return 0.1, dtypes.float32
-  if "conv1d.weight" in name: return 0.1, dtypes.float32
-  if name.endswith("exp_probs_b.bias"): return 0.0, dtypes.float32
-  if name.endswith("norm.weight"): return 1.0, dtypes.bfloat16
-  return 0.001, dtypes.bfloat16
-
-def fake_tp_tensor(shape:tuple[int, ...], value:int|float, dtype, devices:tuple[str, ...], axis:int|None) -> Tensor:
-  if axis is not None and shape[axis] % len(devices): raise ValueError(f"shape {shape} is not TP{len(devices)} divisible on axis {axis}")
-  source = Tensor.full(shape, value, dtype=dtype, device=devices[0]).clone().realize()
-  return source.shard(devices, axis=axis).realize()
-
-def sync(devices:tuple[str, ...]) -> None:
-  for device in devices: Device[device].synchronize()
-
-def main() -> None:
-  parser = argparse.ArgumentParser()
-  parser.add_argument("--devices", type=int, default=8)
-  parser.add_argument("--mode", choices=("attention", "block"), default="attention")
-  parser.add_argument("--iterations", type=int, default=20)
-  args = parser.parse_args()
-  devices = tuple(f"AMD:{i}" for i in range(args.devices))
-  # One exact-width KDA layer, but only 16 fake routed experts. This retains top-k 16 and every
-  # official per-GPU matrix/state shape while keeping fake expert storage below 300 MB per layer.
-  config = replace(kimi_k3_config(4), num_blocks=1, num_experts=16, num_experts_per_tok=16, ssm_layers=(True,),
-                   attn_res_block_size=0)
-  block = GatedDeltaNetBlock(config, config.ssm)
-  begin = time.perf_counter()
-  for name,tensor in nn.state.get_state_dict(block).items():
-    if args.mode == "attention" and name.startswith(("ffn_", "exp_probs_")): continue
-    value, dtype = fake_value(name)
-    tensor.replace(fake_tp_tensor(tuple(int(x) for x in tensor.shape), value, dtype, devices, tp_axis(name)))
-  sync(devices)
-  print(f"fake weights: {time.perf_counter()-begin:.3f}s", flush=True)
-  x_source = (((Tensor.arange(config.dim, dtype=dtypes.float32).reshape(1, 1, config.dim) % 31) / 31) \
-    .cast(dtypes.bfloat16).to(devices[0])).clone().realize()
-  x = x_source.shard(devices, axis=None).realize()
-  block._init_state(x)
-  # Use direct buffer-backed state shards. The production path reaches this form after prefill;
-  # the fake harness begins immediately at decode and must not feed lazy clone graphs to TinyJit.
-  for state,axis in ((block.conv_state_q, 2), (block.conv_state_k, 2), (block.conv_state_v, 2), (block.recurrent_state, 1)):
-    state.replace(Tensor.zeros(*state.shape, dtype=state.dtype, device=devices[0]).shard(devices, axis=axis).realize())
-
-  @TinyJit
-  def run(inp:Tensor) -> Tensor:
-    if args.mode == "attention": return block._attention(block.attn_norm(inp), 0).realize()
-    return block(inp, 0).realize()
-
-  # uncaptured, capture, then replay only
-  run(x); sync(devices)
-  run(x); sync(devices)
-  samples:list[float] = []
-  profile_marker(f"fake K3 {args.mode} start")
-  for _ in range(args.iterations):
-    begin = time.perf_counter(); out = run(x); sync(devices); samples.append((time.perf_counter()-begin)*1e3)
-  profile_marker(f"fake K3 {args.mode} end")
-  print(f"{args.mode}: median={statistics.median(samples):.3f} ms/layer, min={min(samples):.3f} ms/layer, "
-        f"projected_93_layer_rate={1000/(statistics.median(samples)*93):.3f} tok/s, finite={out.float().isfinite().all().item()}")
-
-if __name__ == "__main__": main()

extra/hcq2/ops_amd2.py

@@ -288,15 +288,17 @@ def amd_build_program(prg:UOp) -> UOp:
 
 class AMDAllocator(HCQAllocator['AMDDevice']):
   def __init__(self, dev:AMDDevice):
-    super().__init__(dev, supports_copy_from_disk=dev.has_sdma_queue, supports_transfer=dev.has_sdma_queue and not dev.is_usb())
+    super().__init__(dev, supports_copy_from_disk=dev.has_copy_queue, supports_transfer=dev.has_copy_queue and not dev.is_usb())
 
   def _alloc(self, size:int, options:BufferSpec) -> HCQ2Buffer:
-    return self.dev.iface.alloc(size, host=options.host, uncached=options.uncached, cpu_access=options.cpu_access or not self.dev.has_sdma_queue)
+    return self.dev.iface.alloc(size, host=options.host, uncached=options.uncached, cpu_access=options.cpu_access or not self.dev.has_copy_queue)
 
   def _do_free(self, opaque, options:BufferSpec): self.dev.iface.free(opaque)
 
   def _do_map(self, buf:HCQ2Buffer): return self.dev.iface.map(buf._base if buf._base is not None else buf)
 
+  def _do_unmap(self, buf:HCQ2Buffer): self.dev.iface.unmap(buf)
+
 @dataclass
 class AMDQueueDesc:
   ring: Buffer; read_ptr: Buffer; write_ptr: Buffer; doorbell: Buffer; put_value: Buffer  # noqa: E702
@@ -388,15 +390,24 @@ class KFDIface:
     return hcqbuf
 
   def free(self, mem):
+    self._unmap(mem)
+    if mem.va_addr: FileIOInterface.munmap(mem.va_addr, mem.size)
+    kfd.AMDKFD_IOC_FREE_MEMORY_OF_GPU(self.kfd, handle=mem.meta.handle)
+
+  def unmap(self, mem):
+    self._unmap(mem)
+    if getattr(mem, '_owns_kfd_handle', False): kfd.AMDKFD_IOC_FREE_MEMORY_OF_GPU(self.kfd, handle=mem.meta.handle)
+
+  def _unmap(self, mem):
     gpus = (ctypes.c_int32 * 1)(self.gpu_id)
     stm = kfd.AMDKFD_IOC_UNMAP_MEMORY_FROM_GPU(self.kfd, handle=mem.meta.handle, device_ids_array_ptr=ctypes.addressof(gpus), n_devices=1)
     assert stm.n_success == 1
-    if mem.owner == self.dev:
-      if mem.va_addr: FileIOInterface.munmap(mem.va_addr, mem.size)
-      kfd.AMDKFD_IOC_FREE_MEMORY_OF_GPU(self.kfd, handle=mem.meta.handle)
 
   def map(self, mem):
-    if mem.owner is not None and mem.owner._is_cpu(): return self.alloc(mem.size, host=True, cpu_addr=mem.va_addr)
+    if mem.owner is not None and mem.owner._is_cpu():
+      mapped = self.alloc(mem.size, host=True, cpu_addr=mem.va_addr)
+      mapped._owns_kfd_handle = True
+      return mapped
 
     c_gpus = (ctypes.c_int32 * 1)(self.gpu_id)
     stm = kfd.AMDKFD_IOC_MAP_MEMORY_TO_GPU(self.kfd, handle=mem.meta.handle, device_ids_array_ptr=ctypes.addressof(c_gpus), n_devices=1)
@@ -468,6 +479,7 @@ class PCIIface(PCIIfaceBase):
 
   def require_profile_mode(self): return True
   def is_wgp_active(self, xcc, se, sa, wgp) -> bool: return True # TODO: account for WGP disablement on some asics.
+  def unmap(self, mem): self.free(mem)
 
   def _compute_props(self):
     self.ip_versions = self.dev_impl.ip_ver
@@ -581,7 +593,7 @@ class AMDDevice(HCQ2Compiled):
 
     self.max_copy_size = 0x40000000 if self.iface.ip_versions[am.SDMA0_HWIP][0] >= 5 else 0x400000
     self.sdma_queues:dict = {}
-    self.has_sdma_queue = True # self.sdma_queue(0) is not None, TODO: think of this
+    self.has_copy_queue = not getenv("AMD_DISABLE_SDMA")
 
     super().__init__(device, AMDAllocator(self), [HIPRenderer, AMDLLVMRenderer, HIPCCRenderer], None, can_recover=self.is_am(), arch=self.arch)
 

extra/runbook_digitalocean_mi350x.md

@@ -0,0 +1,235 @@
+# Runbook: Llama 3 8B Training on DigitalOcean MI350X
+
+## Machine Specs
+- 8x MI350X GPUs (gfx950, device ID 75b0), 288GB VRAM each
+- 2TB RAM, 192 CPUs, 2TB disk
+- ROCm 7.14 at `/opt/rocm` (NOT `/opt/rocm-7.1.1` like the submission scripts assume)
+- Python 3.12
+
+## Phase 1: System Setup
+
+### 1.1 Install packages
+```bash
+apt-get update
+apt-get install -y python3-pip python3-venv git tmux rclone clang
+```
+
+### 1.2 Install Python deps
+```bash
+python3 -m pip install --break-system-packages numpy tqdm wandb tiktoken sentencepiece
+```
+
+### 1.3 Install ROCm dev headers
+The base image has ROCm runtime but NOT the HIP dev headers. Need:
+```bash
+apt-get install -y amdrocm-core-dev
+```
+This installs `hip/hip_runtime.h` at `/opt/rocm/core-7.14/include/hip/hip_runtime.h`.
+The symlink `/opt/rocm/include` → `/opt/rocm/core-7.14/include` makes it available at `/opt/rocm/include/hip/hip_runtime.h`.
+
+### 1.4 Configure ROCm comgr
+ROCm 7.14 ships comgr 3.3 at `/opt/rocm/lib/libamd_comgr.so`. tinygrad's DLL loader needs explicit env vars to find it (it searches for `libcomgr.so*` by default, not `libamd_comgr.so*`). Set these in the run command:
+```bash
+export COMGR_PATH=/opt/rocm/lib/libamd_comgr.so
+export COMGR_3_PATH=/opt/rocm/lib/libamd_comgr.so
+```
+Also add ROCm libs to ldconfig so comgr's shared library dependencies resolve:
+```bash
+cat > /etc/ld.so.conf.d/rocm.conf << 'EOF'
+/opt/rocm/lib
+/opt/rocm/lib/llvm/lib
+/opt/rocm/lib/rocm_sysdeps/lib
+EOF
+ldconfig
+```
+
+### 1.5 Install geohot tmux config
+```bash
+curl -sL https://raw.githubusercontent.com/geohot/configuration/master/.tmux.conf -o ~/.tmux.conf
+```
+
+### 1.6 Reload amdgpu driver
+tinygrad's HCQ backend needs `/dev/kfd` which is created by the amdgpu kernel driver.
+If the driver was unloaded, reload it:
+```bash
+modprobe amdgpu
+ls /dev/kfd  # should exist
+```
+
+## Phase 2: Clone tinygrad
+```bash
+cd /root
+git clone https://github.com/tinygrad/tinygrad.git
+cd tinygrad
+python3 -m pip install --break-system-packages -e .
+```
+
+## Phase 3: Download C4 Dataset
+
+The C4 data is on the MLCommons Cloudflare R2 bucket in Megatron-LM indexed format.
+
+```bash
+rclone config create mlc-training s3 provider=Cloudflare \
+  access_key_id=76ea42eadb867e854061a1806220ee1e \
+  secret_access_key=a53625c4d45e3ca8ac0df8a353ea3a41ffc3292aa25259addd8b7dc5a6ce2936 \
+  endpoint=c2686074cb2caf5cbaf6d134bdba8b47.r2.cloudflarestorage.com
+
+mkdir -p /root/datasets/c4-8b
+rclone copy mlc-training:mlcommons-training-wg-public/llama3_1/datasets/c4/llama3_1_8b/ /root/datasets/c4-8b/ -P
+```
+
+Files downloaded (~85GB total, ~6 minutes):
+- `c4-train.en_6_text_document.bin` (79 GB)
+- `c4-train.en_6_text_document.idx` (870 MB)
+- `c4-validation-91205-samples.en_text_document.bin` (159 MB)
+- `c4-validation-91205-samples.en_text_document.idx` (1.8 MB)
+- `LICENSE.txt`, `NOTICE.txt`
+
+### Symlink for the submission script
+The `dev_run.sh` script hardcodes `BASEDIR="/raid/datasets/c4-8b/"`. Symlink:
+```bash
+mkdir -p /raid/datasets
+ln -s /root/datasets/c4-8b /raid/datasets/c4-8b
+```
+
+## Phase 4: wandb Login
+```bash
+wandb login
+```
+Enter API key from https://wandb.ai/authorize
+
+## Phase 5: Run Training
+
+### 5.1 Smoke test (beam search, 2 layers, fake data)
+Always run beam first to validate the pipeline:
+```bash
+cd /root/tinygrad
+COMGR_PATH=/opt/rocm/lib/libamd_comgr.so \
+COMGR_3_PATH=/opt/rocm/lib/libamd_comgr.so \
+CC=/opt/rocm/core-7.14/lib/llvm/bin/clang \
+DEV=AMD:HIP \
+ROCM_PATH=/opt/rocm BASEDIR=/root/datasets/c4-8b/ \
+  bash examples/mlperf/training_submission_v6.0/tinycorp/benchmarks/llama31_8b/implementations/tinybox_8xMI350X/dev_beam.sh
+```
+
+### 5.2 Full training run
+```bash
+cd /root/tinygrad
+COMGR_PATH=/opt/rocm/lib/libamd_comgr.so \
+COMGR_3_PATH=/opt/rocm/lib/libamd_comgr.so \
+CC=/opt/rocm/core-7.14/lib/llvm/bin/clang \
+DEV=AMD:HIP \
+ROCM_PATH=/opt/rocm BASEDIR=/root/datasets/c4-8b/ \
+WANDB=1 \
+  bash examples/mlperf/training_submission_v6.0/tinycorp/benchmarks/llama31_8b/implementations/tinybox_8xMI350X/dev_run.sh
+```
+
+## Environment Variable Reference
+
+| Variable | Value | Why |
+|---|---|---|
+| `COMGR_PATH` | `/opt/rocm/lib/libamd_comgr.so` | tinygrad's DLL loader needs explicit path to find comgr 3.3 |
+| `COMGR_3_PATH` | `/opt/rocm/lib/libamd_comgr.so` | comgr 3.x uses a separate `comgr_3` module with its own path var |
+| `CC` | `/opt/rocm/core-7.14/lib/llvm/bin/clang` | System clang doesn't know gfx950; must use ROCm's bundled clang |
+| `DEV` | `AMD:HIP` | Force HIPRenderer (comgr-based) over HIPCCRenderer (hipcc subprocess) |
+| `ROCM_PATH` | `/opt/rocm` | Script defaults to `/opt/rocm-7.1.1` which doesn't exist |
+| `BASEDIR` | `/root/datasets/c4-8b/` | Where C4 dataset was downloaded (script hardcodes `/raid/datasets/c4-8b/`) |
+| `WANDB` | `1` | Enable wandb logging (off by default) |
+
+## Architecture
+
+| Component | Source file |
+|---|---|
+| Model | `examples/mlperf/models/flat_llama.py` — FlatTransformer, FP8 MXFP4 weights, fused QKV, flash attention |
+| Trainer | `examples/mlperf/model_train.py` → `train_llama3()` |
+| Optimizer | `examples/mlperf/optim.py` — GradAccClipAdamW, master weights, FP8 re-quant |
+| LR schedule | `examples/mlperf/lr_schedulers.py` — CosineAnnealingLRWithWarmup |
+| Dataloader | `examples/mlperf/dataloader.py` — Megatron-LM indexed bin format |
+| ASM GEMM | `extra/gemm/cdna_asm_gemm.py` — gfx950 MFMA assembly, MXFP4 |
+| Flash attention | `extra/thunder/amd/fa.py` |
+| Fused kernels | `extra/llama_kernels/` — rmsnorm, silu, quantize, fused_ce |
+| GPU driver | `tinygrad/runtime/ops_amd.py` — HCQ, direct KFD ioctl |
+| Renderer | `tinygrad/renderer/cstyle.py` — HIPRenderer for gfx950 |
+| comgr compiler | `tinygrad/runtime/support/compiler_amd.py` — HIPCompiler using comgr 3.3 |
+
+## Troubleshooting
+
+### `'hip/hip_runtime.h' file not found`
+Install `amdrocm-core-dev`:
+```bash
+apt-get install -y amdrocm-core-dev
+```
+
+### `'gfx950' is not a recognized processor` + LLVM crash
+System clang doesn't know gfx950. Set `CC=/opt/rocm/core-7.14/lib/llvm/bin/clang`.
+
+### `comgr not available: try setting COMGR_PATH?`
+Add ROCm libs to ldconfig and set `COMGR_PATH` and `COMGR_3_PATH`:
+```bash
+# /etc/ld.so.conf.d/rocm.conf should contain /opt/rocm/lib paths
+ldconfig
+```
+
+### `comgr not available: try setting COMGR_3_PATH?`
+comgr 3.x uses a separate module. Set `COMGR_3_PATH=/opt/rocm/lib/libamd_comgr.so` too.
+
+### `FileNotFoundError: '/raid/datasets/c4-8b/...'`
+Script hardcodes `BASEDIR`. Either symlink or edit the script:
+```bash
+mkdir -p /raid/datasets && ln -s /root/datasets/c4-8b /raid/datasets/c4-8b
+```
+
+### `No such file or directory: 'clang'`
+Install clang: `apt-get install -y clang` (for CPU compilation).
+For gfx950 HIP compilation, comgr (not clang) is used — ensure the ROCm 7.14 comgr 3.3 is properly loaded via `COMGR_PATH` and `COMGR_3_PATH`.
+
+## Appendix: KVM Virtualization Observations
+
+### Virtualization detection
+```
+$ systemd-detect-virt
+kvm
+$ lspci -nn | grep AMD
+83:00.0 ... Device [1002:75b0]
+```
+CPU flags include `hypervisor`. `dmesg` shows `Hypervisor detected: KVM`.
+
+### PCI device ID
+`lspci -v` shows device ID `0x75b0` and subsystem ID `0x75a0`:
+```
+83:00.0 Processing accelerators: ... Device 75b0
+    Subsystem: ... Device 75a0
+```
+tinygrad's `PCIIface` in `ops_amd.py` and `hive_reset.py` did not list `0x75b0`, so the GPU was not found. Adding `0x75b0` to the device ID list in both files fixes the detection.
+
+### amdgpu driver behavior
+On first boot, amdgpu loaded and bound to all 8 GPUs. On one boot it failed to initialize:
+```
+[  799.780369] amdgpu 0000:83:00.0: Failed to alloc msi vectors
+[  799.781476] amdgpu 0000:83:00.0: sw_init of IP block <vega20_ih> failed -22
+[  799.782724] amdgpu 0000:83:00.0: amdgpu_device_ip_init failed
+[  799.793885] amdgpu 0000:83:00.0: Fatal error during GPU init
+```
+On a subsequent boot, amdgpu initialized successfully (SMU initialized, VRAM ready). After unbinding all 8 GPUs from amdgpu, `rmmod amdgpu` wedged the module (stuck in "Unloading" state in `/proc/modules`), requiring a full VM reboot.
+
+### `/dev/kfd`
+`/dev/kfd` exists when amdgpu is loaded. Opening it returns `OSError: [Errno 22] Invalid argument`.
+
+### VRAM BAR reads all 0xFF
+After amdgpu initializes the GPU and is then unbound, reading the VRAM BAR (via `/sys/bus/pci/devices/0000:83:00.0/resource0`) returns all `0xFF` at all offsets — including the discovery table at `vram_size - 64KB`. tinygrad's `AMDev._run_discovery()` fails with `AssertionError: discovery signatures mismatch`.
+
+A PCI reset (`echo 1 > /sys/bus/pci/devices/0000:83:00.0/reset`) did not change the VRAM contents — still all `0xFF`.
+
+VRAM was also all `0xFF` when read via `/dev/mem` at the BAR physical address (`0xa0000000000`).
+
+### VFIO attempt
+Bound the GPU to `vfio-pci` with `enable_unsafe_noiommu_mode=1`. The GPU bound successfully and `/dev/vfio/noiommu-0` appeared. Running tinygrad with `VFIO=1` still failed with the same `discovery signatures mismatch` — VRAM BAR still reads all `0xFF`.
+
+### No IOMMU in guest
+`dmesg` has no `AMD-Vi` entries. PCI devices have no `iommu_group` symlink.
+
+### No fan control
+No `fan*` or `pwm*` hwmon entries exist. Only `temp*`, `power*`, `freq*` are exposed. GPU temps read 56-63°C, power ~265W per GPU.
+
+### Current status: NOT WORKING
+tinygrad's `PCIIface` finds the GPU (after adding `0x75b0`) but `AMDev._run_discovery()` fails because the VRAM discovery table reads all `0xFF`. This was observed with the GPU unbound from any driver, after PCI reset, and with VFIO bound.

extra/thunder/amd/matvec_bf16_splitk.cpp

@@ -1,74 +0,0 @@
-#include "kittens.cuh"
-
-using namespace kittens;
-
-#ifndef MATVEC_N
-#define MATVEC_N 1536
-#endif
-#ifndef MATVEC_K
-#define MATVEC_K 7168
-#endif
-
-constexpr int SPLIT_WAVES = 8;
-
-template<int W>
-__device__ __forceinline__ float run_split(const bf16 *A_ptr, const bf16 *B_ptr, int out_base,
-                                           st_bf<16, 32, st_16x32_s> &As,
-                                           st_bf<16, 32, st_16x32_s> &Bs) {
-  constexpr int K = MATVEC_K;
-  rt_bf<16, 32, row_l, rt_16x32_s> A;
-  rt_bf<16, 32, row_l, rt_16x32_s> B;
-  rt_fl<16, 16, col_l, rt_16x16_s> C;
-  zero(C);
-  const int lane = laneid();
-  constexpr int k_begin = W * (K / SPLIT_WAVES), k_end = k_begin + K / SPLIT_WAVES;
-  #pragma unroll 1
-  for (int k = k_begin; k < k_end; k += 32) {
-    #pragma unroll
-    for (int idx = lane; idx < 16 * 32; idx += 64) {
-      const int row = idx / 32, col = idx % 32;
-      *reinterpret_cast<bf16 *>(reinterpret_cast<char *>(&As.data[0]) + As.swizzle({row, col})) = A_ptr[k + col];
-      *reinterpret_cast<bf16 *>(reinterpret_cast<char *>(&Bs.data[0]) + Bs.swizzle({row, col})) =
-        B_ptr[(out_base + row) * K + k + col];
-    }
-    asm volatile("s_waitcnt lgkmcnt(0)");
-    load(A, As);
-    load(B, Bs);
-    asm volatile("s_waitcnt lgkmcnt(0)");
-    mma_ABt(C, A, B, C);
-  }
-  return C.tiles[0][0].data[0].x;
-}
-
-// Eight waves split K for one 16-channel output tile.  Each wave uses MFMA on
-// a repeated activation row, then wave zero reduces the eight FP32 partials.
-__global__ __launch_bounds__(64 * SPLIT_WAVES, 1)
-void hk_bf16_matvec_splitk(bf16 *C_ptr, const bf16 *A_ptr, const bf16 *B_ptr, bf16 *unused) {
-  constexpr int N = MATVEC_N, K = MATVEC_K;
-  static_assert(N % 16 == 0 && K % (32 * SPLIT_WAVES) == 0);
-  __shared__ st_bf<16, 32, st_16x32_s> As[SPLIT_WAVES];
-  __shared__ st_bf<16, 32, st_16x32_s> Bs[SPLIT_WAVES];
-  __shared__ float partial[SPLIT_WAVES][16];
-  const int tid = threadIdx.x, wave = tid / 64, lane = tid & 63;
-  const int out_base = blockIdx.x * 16;
-  float result = 0.0f;
-  switch (wave) {
-    case 0: result = run_split<0>(A_ptr, B_ptr, out_base, As[0], Bs[0]); break;
-    case 1: result = run_split<1>(A_ptr, B_ptr, out_base, As[1], Bs[1]); break;
-    case 2: result = run_split<2>(A_ptr, B_ptr, out_base, As[2], Bs[2]); break;
-    case 3: result = run_split<3>(A_ptr, B_ptr, out_base, As[3], Bs[3]); break;
-    case 4: result = run_split<4>(A_ptr, B_ptr, out_base, As[4], Bs[4]); break;
-    case 5: result = run_split<5>(A_ptr, B_ptr, out_base, As[5], Bs[5]); break;
-    case 6: result = run_split<6>(A_ptr, B_ptr, out_base, As[6], Bs[6]); break;
-    case 7: result = run_split<7>(A_ptr, B_ptr, out_base, As[7], Bs[7]); break;
-  }
-  if (lane < 16) partial[wave][lane] = result;
-  asm volatile("s_waitcnt lgkmcnt(0)");
-  __builtin_amdgcn_s_barrier();
-  if (wave == 0 && lane < 16) {
-    float total = 0.0f;
-    #pragma unroll
-    for (int i = 0; i < SPLIT_WAVES; i++) total += partial[i][lane];
-    C_ptr[out_base + lane] = static_cast<bf16>(total);
-  }
-}

test/backend/test_linearizer.py

@@ -437,7 +437,7 @@ def reset_bufs(bufs:list[Buffer]):
   for buf in bufs: buf.copy_from(Buffer("PYTHON", buf.size, buf.dtype, opaque=memoryview(bytearray(buf.nbytes))))
 
 def _helper_linearizer_opt_ast(realized_ast:UOp, real_bufs:list[Buffer], opts=[],
-                               apply_tc=False, atol=1e-4, rtol=1e-4, color_sizes=[], wanna_output=[]):
+                               apply_tc=False, atol=1e-4, rtol=1e-4, color_sizes=[], wanna_output=[], check_default_opt=True):
   outbufs = real_bufs[:len(realized_ast.src)]
   wanna_output = [np.array(x).flatten() for x in wanna_output]
   buf_uops = [UOp.new_buffer(b.device, b.size, b.dtype) for b in real_bufs]
@@ -459,9 +459,7 @@ def _helper_linearizer_opt_ast(realized_ast:UOp, real_bufs:list[Buffer], opts=[]
     for buf,want in zip(copyout_outputs(outbufs), wanna_output): np.testing.assert_allclose(buf, want, atol=atol, rtol=rtol)
 
   # Check correctness of handcoded optimiztions.
-  reset_bufs(outbufs)
-  run_prg(opts=None)
-  for buf,want in zip(copyout_outputs(outbufs), wanna_output): np.testing.assert_allclose(buf, want, atol=atol, rtol=rtol)
+  if check_default_opt: check_opt(None)
   for x in opts: # Check custom transformations if any.
     check_opt(([Opt(OptOps.TC, 0, (TC_SELECT.value, TC_OPT.value, 1))] if apply_tc else [])+x)
 

test/device/test_hcq2.py

@@ -0,0 +1,14 @@
+import unittest, numpy as np
+from unittest.mock import patch
+from tinygrad import Device, Tensor
+from tinygrad.helpers import getenv
+from tinygrad.runtime.support.hcq2 import HCQ_DEVS, all_devices_in
+
+@unittest.skipUnless(getenv("HCQ2") and all_devices_in(Device.DEFAULT, HCQ_DEVS), "hcq2 device required")
+class TestHCQ2(unittest.TestCase):
+  def test_copy_without_copy_queue(self):
+    with patch.object(Device[Device.DEFAULT], "has_copy_queue", False):
+      np.testing.assert_equal(Tensor(np.arange(61, dtype=np.float32)).to(Device.DEFAULT).contiguous().realize().numpy(), np.arange(61))
+
+if __name__ == "__main__":
+  unittest.main()

test/helpers.py

@@ -48,6 +48,8 @@ def check_schedule(t:Tensor|list[Tensor]|UOp, allowed:int, to_prerealize:list[Te
   else:
     assert isinstance(t, UOp), f"can't schedule {t}"
     linear, var_vals = Tensor(t).linear_with_vars()
+  # test compiling the linear
+  compile_linear(linear)
   kernel_cnt = sum((len(call.device) if isinstance(call.device, tuple) else 1)
                    for call in linear.src if call.src[0].op is Ops.SINK or not filter_sink)
   if kernel_cnt != allowed:
@@ -57,8 +59,6 @@ def check_schedule(t:Tensor|list[Tensor]|UOp, allowed:int, to_prerealize:list[Te
         print("kernel", i+1)
         print(call.src[0])
     raise KernelCountException(allowed, kernel_cnt)
-  # test compiling the linear
-  compile_linear(linear)
   return linear, var_vals
 
 def assert_kernel_count(expected:int):

test/null/test_kimi.py

@@ -1,30 +0,0 @@
-import unittest
-from tinygrad import Tensor, dtypes, nn
-from tinygrad.llm.kimi import _shard_kimi
-from tinygrad.llm.model import SSMConfig, Transformer, TransformerConfig
-
-class TestKimiTP4(unittest.TestCase):
-  def test_prefill_and_decode_graph(self):
-    devices = ("NULL:0", "NULL:1", "NULL:2", "NULL:3")
-    config = TransformerConfig(num_blocks=2, dim=32, hidden_dim=128, n_heads=4, n_kv_heads=1, norm_eps=1e-5,
-      vocab_size=64, head_dim=12, rope_theta=10000, rope_dim=4, v_head_dim=8, max_context=4, kv_lora_rank=16,
-      num_experts=8, num_experts_per_tok=2, norm_topk_prob=True, shared_expert_dim=32, ssm_layers=(True, False),
-      ssm=SSMConfig(4, 8, 4, 4, 32, True), shared_expert_gate=False, leading_dense_blocks=1, dense_hidden_dim=64,
-      routed_scaling_factor=2.446, expert_bias=True, expert_mxfp4=True, bf16_activations=True, kda_split_qkv=True)
-    model = Transformer(config)
-    for name, value in nn.state.get_state_dict(model).items():
-      fill = 127 if name.endswith("weight_scale") else 0
-      value.replace(Tensor.full(value.shape, fill, dtype=value.dtype if value.dtype is dtypes.uint8 else dtypes.bfloat16, device="NULL"))
-    _shard_kimi(model, devices)
-
-    temperature = Tensor([0.0], device=devices)
-    prefill = model(Tensor([[1, 2]], dtype=dtypes.int32, device=devices), 0, temperature).realize()
-    model(Tensor([[1, 2]], dtype=dtypes.int32, device=devices), 0, temperature).realize()  # replay prefill JIT
-    decode = model(Tensor([[3]], dtype=dtypes.int32, device=devices), 2, temperature).realize()
-    model(Tensor([[4]], dtype=dtypes.int32, device=devices), 3, temperature).realize()  # replay decode JIT
-    self.assertEqual(prefill.shape, (1, 1))
-    self.assertEqual(decode.shape, (1, 1))
-    self.assertEqual(model.blk[0].recurrent_state.uop.axis, 1)
-    self.assertEqual(model.blk[1].cache_k.dtype, dtypes.bfloat16)
-
-if __name__ == "__main__": unittest.main()

test/null/test_kimi_k3.py

@@ -1,29 +0,0 @@
-import unittest
-from tinygrad import Tensor, dtypes, nn
-from tinygrad.llm.kimi_k3 import _shard_kimi_k3
-from test.unit.test_llm_k3 import small_k3_config
-from tinygrad.llm.model import Transformer
-
-class TestKimiK3TP8(unittest.TestCase):
-  @staticmethod
-  def _model():
-    model = Transformer(small_k3_config())
-    for name,value in nn.state.get_state_dict(model).items():
-      fill = 127 if name.endswith("weight_scale") else 0
-      dtype = value.dtype if value.dtype is dtypes.uint8 else dtypes.bfloat16
-      value.replace(Tensor.full(value.shape, fill, dtype=dtype, device="NULL"))
-    _shard_kimi_k3(model, tuple(f"NULL:{i}" for i in range(8)))
-    return model
-
-  def test_prefill_decode_and_jit_replay(self):
-    devices = tuple(f"NULL:{i}" for i in range(8))
-    model = self._model()
-    temperature = Tensor([0.0], device=devices)
-    self.assertEqual(model(Tensor([[1, 2]], dtype=dtypes.int32, device=devices), 0, temperature).realize().shape, (1, 1))
-    model(Tensor([[1, 2]], dtype=dtypes.int32, device=devices), 0, temperature).realize()
-    self.assertEqual(model(Tensor([[3]], dtype=dtypes.int32, device=devices), 2, temperature).realize().shape, (1, 1))
-    model(Tensor([[4]], dtype=dtypes.int32, device=devices), 3, temperature).realize()
-    self.assertEqual(model.blk[0].recurrent_state.uop.axis, 1)
-    self.assertEqual(model.blk[1].cache_k.dtype, dtypes.bfloat16)
-
-if __name__ == "__main__": unittest.main()

test/null/test_memory_planner.py

@@ -27,10 +27,15 @@ def _make_linear(buffer_lists, copies=None):
     calls.append(UOp(Ops.CALL, src=(src0, *bufs)))
   return UOp(Ops.LINEAR, src=tuple(calls))
 
+def _get_planned_view(buf:UOp) -> tuple[UOp, int, int]|None:
+  view = buf.src[0] if buf.op is Ops.BITCAST else buf
+  if view.op is not Ops.SHRINK or view.src[0].op is not Ops.BUFFER: return None
+  return (arena:=view.src[0]), view.src[1].val * arena.dtype.itemsize, view.src[2].val * arena.dtype.itemsize
+
 def _get_arena(buf, linear, result):
   for orig_si, new_si in zip(linear.src, result.src):
     for orig, new in zip(orig_si.src[1:], new_si.src[1:]):
-      if orig is buf and new.op is Ops.SLICE: return new.src[0]
+      if orig is buf and (planned:=_get_planned_view(new)) is not None: return planned[0]
   return None
 
 def check_assign(buffer_lists, copies=None):
@@ -41,8 +46,8 @@ def check_assign(buffer_lists, copies=None):
   replace_map: dict[int, tuple[UOp, int, int]] = {}
   for orig_si, new_si in zip(linear.src, result.src):
     for orig, new in zip(orig_si.src[1:], new_si.src[1:]):
-      if new.op is Ops.SLICE and id(orig) not in replace_map:
-        replace_map[id(orig)] = (new.src[0], new.src[1].val * new.src[0].dtype.itemsize, new.arg * new.dtype.itemsize)
+      if (planned:=_get_planned_view(new)) is not None and id(orig) not in replace_map:
+        replace_map[id(orig)] = planned
 
   # verify pinned buffers are not planned
   for buf in held_bufs:

test/opt/test_tensor_cores.py

@@ -79,7 +79,8 @@ class TestTensorCores(unittest.TestCase):
   @unittest.skipUnless(Device[Device.DEFAULT].renderer.tensor_cores, "test requires tensor cores")
   def test_tensor_cores(self):
     for tc in Device[Device.DEFAULT].renderer.tensor_cores:
-      helper_tc_allclose(tc.dims[0], tc.dims[1], tc.dims[2], tc.dtype_in, tc.dtype_out, axis=0, tc_opt=0)
+      with self.subTest(tc=tc):
+        helper_tc_allclose(tc.dims[0], tc.dims[1], tc.dims[2], tc.dtype_in, tc.dtype_out, axis=0, tc_opt=0)
 
   @unittest.skipUnless(Device[Device.DEFAULT].renderer.tensor_cores, "test requires tensor cores")
   def test_tensor_cores_nested_reduce(self):
@@ -185,10 +186,10 @@ class TestTensorCores(unittest.TestCase):
     # skip fp8 tcs: the unoptimized ALU baseline quantizes products to fp8 (JAX promotion), which legitimately
     # differs from the MFMA path (f32 accumulation), so the baseline-vs-TC numerical gate can't hold for fp8.
     tc = next(tc for tc in Device[Device.DEFAULT].renderer.tensor_cores if tc.dtype_in not in dtypes.fp8s)
-    x, y = Tensor.rand(64, 64, dtype=tc.dtype_in), Tensor.rand(64, 64, dtype=tc.dtype_in)
+    x, y = Tensor.rand(16, 64, dtype=tc.dtype_in), Tensor.rand(64, 16, dtype=tc.dtype_in)
     r = x.matmul(y, dtype=tc.dtype_out)
     opts = [Opt(OptOps.UNROLL, 0, 2)]
-    ast = helper_linearizer_opt(r, [opts], apply_tc=True, atol=3e-2, rtol=1e-3)
+    ast = helper_linearizer_opt(r, [opts], apply_tc=True, atol=3e-2, rtol=1e-3, check_default_opt=False)
     for u in tuple(to_program(replace_opts(ast, opts), Device[Device.DEFAULT].renderer).src[1].src):
       if u.op is Ops.WMMA:
         assert u.src[-1].src[0].op != Ops.STORE
@@ -199,10 +200,10 @@ class TestTensorCores(unittest.TestCase):
   @unittest.skipIf(Device.DEFAULT in {"CPU"}, "CPU does not support using a different type for accumulation")
   def test_tensor_cores_unroll_casted_phi(self):
     tc = [tc for tc in Device[Device.DEFAULT].renderer.tensor_cores if tc.dtype_in != tc.dtype_out and tc.dtype_in not in dtypes.fp8s][0]
-    x, y = Tensor.rand(64, 64, dtype=tc.dtype_in), Tensor.rand(64, 64, dtype=tc.dtype_in)
+    x, y = Tensor.rand(16, 64, dtype=tc.dtype_in), Tensor.rand(64, 16, dtype=tc.dtype_in)
     r = x.matmul(y, dtype=tc.dtype_out)
     opts = [Opt(OptOps.UNROLL, 0, 2)]
-    ast = helper_linearizer_opt(r, [opts], apply_tc=True, atol=3e-2, rtol=1e-3)
+    ast = helper_linearizer_opt(r, [opts], apply_tc=True, atol=3e-2, rtol=1e-3, check_default_opt=False)
     for u in tuple(to_program(replace_opts(ast, opts), Device[Device.DEFAULT].renderer).src[1].src):
       if u.op is Ops.WMMA:
         #assert u.src[-1].dtype == dtypes.float.vec(prod(tc.thread_local_sizes[2]))
@@ -215,10 +216,10 @@ class TestTensorCores(unittest.TestCase):
   def test_tensor_cores_unroll_casted_phi_with_children(self):
     # all STORE children are outside the loop
     tc = [tc for tc in Device[Device.DEFAULT].renderer.tensor_cores if tc.dtype_in != tc.dtype_out and tc.dtype_in not in dtypes.fp8s][0]
-    x, y = Tensor.rand(64, 64, dtype=tc.dtype_in), Tensor.rand(64, 64, dtype=tc.dtype_in)
+    x, y = Tensor.rand(16, 64, dtype=tc.dtype_in), Tensor.rand(64, 16, dtype=tc.dtype_in)
     r = x.matmul(y, dtype=tc.dtype_out).relu()
     opts = [Opt(OptOps.UNROLL, 0, 2)]
-    ast = helper_linearizer_opt(r, [opts], apply_tc=True, atol=3e-2, rtol=1e-3)
+    ast = helper_linearizer_opt(r, [opts], apply_tc=True, atol=3e-2, rtol=1e-3, check_default_opt=False)
     for u in tuple(to_program(replace_opts(ast, opts), Device[Device.DEFAULT].renderer).src[1].src):
       if u.op is Ops.WMMA:
         #assert u.src[-1].dtype == dtypes.float.vec(prod(tc.thread_local_sizes[2]))

test/unit/test_attention.py

@@ -1,11 +1,9 @@
 import unittest
-from types import SimpleNamespace
 import numpy as np
-from tinygrad import Tensor, dtypes, nn
-from tinygrad.llm.kimi import _shard_kimi
+from tinygrad import Tensor, dtypes
 from tinygrad.llm.model import (
   GatedDeltaNetBlock, SSMConfig, TransformerBlock, TransformerConfig,
-  apply_rope as apply_rope_new, iterative_topk, l2norm, precompute_freqs_cis, pairwise_topk,
+  apply_rope as apply_rope_new, precompute_freqs_cis, pairwise_topk,
 )
 
 def apply_rope(x:Tensor, start_pos:int):
@@ -43,11 +41,6 @@ class TestAttention(unittest.TestCase):
     np.testing.assert_allclose(block.cache_kv[0, :, :, :seqlen, :].numpy(), expected.numpy(), rtol=1e-5, atol=1e-5)
 
 class TestGatedDeltaNetBlock(unittest.TestCase):
-  def test_kda_l2norm_matches_fla(self):
-    x = np.array([[1e-4, -2e-4, 3e-4], [1.0, 2.0, -3.0]], dtype=np.float32)
-    expected = x / np.sqrt((x*x).sum(axis=-1, keepdims=True) + 1e-6)
-    np.testing.assert_allclose(l2norm(Tensor(x)).numpy(), expected, rtol=1e-6, atol=1e-6)
-
   def _tensor_linspace(self, start:float, stop:float, shape:tuple[int, ...]) -> Tensor:
     return Tensor.linspace(start, stop, int(np.prod(shape)), dtype=dtypes.float32).reshape(*shape)
 
@@ -197,87 +190,6 @@ class TestGatedDeltaNetBlock(unittest.TestCase):
     alpha = np.exp(-self._softplus_np(np.arange(1, 5)).reshape(1, 2, 1, 2))
     np.testing.assert_allclose(block.recurrent_state.numpy(), initial_state.numpy() * alpha, rtol=1e-5, atol=1e-5)
 
-  def test_kda_safe_gate_decay(self):
-    config = self._make_config(n_heads=2, kda_full_rank_gate=True, kda_gate_lower_bound=-5.0,
-      ssm=SSMConfig(conv_kernel=2, state_size=2, group_count=2, time_step_rank=2, inner_size=4, kda=True))
-    block, x = GatedDeltaNetBlock(config, config.ssm), Tensor([[[1., 2., 0., 0.]]])
-    block.ssm_f_a.weight = Tensor([[1., 0., 0., 0.], [0., 1., 0., 0.]])
-    block.ssm_f_b.weight = Tensor([[1., 0.], [0., 1.], [1., 1.], [2., 1.]])
-    block.ssm_dt["bias"] = Tensor.zeros(4)
-    block.ssm_a = Tensor([[-2.], [-3.]])  # stores -exp(A_log)
-    block._init_state(x)
-    initial_state = Tensor.arange(8, dtype=dtypes.float32).reshape(1, 2, 2, 2)
-    block.recurrent_state.assign(initial_state).realize()
-    block._attention(x, 0).realize()
-    gate_logits = np.arange(1, 5, dtype=np.float32).reshape(1, 2, 2)
-    exp_a = np.array([2., 3.], dtype=np.float32).reshape(1, 2, 1)
-    alpha = np.exp(-5.0 / (1.0 + np.exp(-(exp_a * gate_logits)))).reshape(1, 2, 1, 2)
-    np.testing.assert_allclose(block.recurrent_state.numpy(), initial_state.numpy() * alpha, rtol=2e-5, atol=2e-5)
-
-  def test_kda_per_channel_a(self):
-    config = self._make_config(n_heads=2, kda_full_rank_gate=True, kda_gate_lower_bound=-5.0,
-      ssm=SSMConfig(conv_kernel=2, state_size=2, group_count=2, time_step_rank=2, inner_size=4, kda=True, channel_decay=True))
-    block, x = GatedDeltaNetBlock(config, config.ssm), Tensor([[[1., 2., 0., 0.]]])
-    block.ssm_f_a.weight = Tensor([[1., 0., 0., 0.], [0., 1., 0., 0.]])
-    block.ssm_f_b.weight = Tensor([[1., 0.], [0., 1.], [1., 1.], [2., 1.]])
-    block.ssm_dt["bias"] = Tensor.zeros(4)
-    block.ssm_a = Tensor([[-2.], [-3.]])
-    block._init_state(x)
-    initial_state = Tensor.arange(8, dtype=dtypes.float32).reshape(1, 2, 2, 2)
-    block.recurrent_state.assign(initial_state).realize()
-    block._attention(x, 0).realize()
-    gate_logits = np.arange(1, 5, dtype=np.float32).reshape(1, 2, 2)
-    exp_a = np.array([2., 3.], dtype=np.float32).reshape(1, 1, 2)
-    alpha = np.exp(-5.0 / (1.0 + np.exp(-(exp_a * gate_logits)))).reshape(1, 2, 1, 2)
-    np.testing.assert_allclose(block.recurrent_state.numpy(), initial_state.numpy() * alpha, rtol=2e-5, atol=2e-5)
-
-  def test_kda_chunked_prefill_matches_decode(self):
-    config = self._make_config(max_context=4, n_heads=2,
-      ssm=SSMConfig(conv_kernel=2, state_size=2, group_count=2, time_step_rank=2, inner_size=4, kda=True), kda_split_qkv=True)
-    x = Tensor.linspace(-1, 1, 4*config.dim, dtype=dtypes.float32).reshape(1, 4, config.dim).cast(dtypes.bfloat16)
-
-    chunked = GatedDeltaNetBlock(config, config.ssm)
-    sequential = GatedDeltaNetBlock(config, config.ssm)
-    for value in nn.state.get_state_dict(chunked).values(): value.replace(value.cast(dtypes.bfloat16).realize())
-    sequential_state = nn.state.get_state_dict(sequential)
-    for name, value in nn.state.get_state_dict(chunked).items(): sequential_state[name].replace(value)
-
-    chunked._init_state(x)
-    chunk_out = chunked._attention(x, 0).realize()
-    sequential._init_state(x)
-    seq_out = Tensor.cat(*[sequential._attention(x[:, t:t+1], t).realize() for t in range(x.shape[1])], dim=1).realize()
-
-    np.testing.assert_allclose(chunk_out.numpy(), seq_out.numpy(), rtol=1e-5, atol=1e-5)
-    for name in ("conv_state_q", "conv_state_k", "conv_state_v"):
-      np.testing.assert_allclose(getattr(chunked, name).numpy(), getattr(sequential, name).numpy(), rtol=2e-2, atol=4e-3)
-    np.testing.assert_allclose(chunked.recurrent_state.numpy(), sequential.recurrent_state.numpy(), rtol=2e-3, atol=2e-3)
-
-  def test_kda_tp_final_token_matches_unsharded(self):
-    config = self._make_config(dim=8, hidden_dim=16, n_heads=4, n_kv_heads=4, head_dim=2, rope_dim=2, v_head_dim=2,
-      ssm=SSMConfig(conv_kernel=2, state_size=2, group_count=4, time_step_rank=4, inner_size=8, kda=True), kda_split_qkv=True)
-    single, tp = GatedDeltaNetBlock(config, config.ssm), GatedDeltaNetBlock(config, config.ssm)
-    for name, value in nn.state.get_state_dict(single).items():
-      data = np.full(value.shape, 1.0, np.float32) if "norm.weight" in name else \
-             np.linspace(-0.2, 0.2, value.numel(), dtype=np.float32).reshape(value.shape)
-      if name == "ssm_a": data.fill(-0.1)
-      value.replace(Tensor(data, device="CPU", dtype=dtypes.bfloat16).realize())
-    tp_state = nn.state.get_state_dict(tp)
-    for name, value in nn.state.get_state_dict(single).items(): tp_state[name].replace(value)
-    devices = ("CPU", "CPU:1")
-    _shard_kimi(SimpleNamespace(blk=[tp]), devices)
-
-    x = Tensor(np.linspace(-1, 1, 32, dtype=np.float32).reshape(1, 4, 8), device="CPU", dtype=dtypes.bfloat16)
-    single._init_state(x)
-    expected = single._attention(x, 0).realize()
-    x_tp = x.shard(devices, axis=None)
-    tp._init_state(x_tp)
-    actual = tp._attention(x_tp, 0).realize()
-
-    np.testing.assert_equal(actual.numpy(), expected.numpy())
-    np.testing.assert_equal(tp.recurrent_state.numpy(), single.recurrent_state.numpy())
-    for name in ("conv_state_q", "conv_state_k", "conv_state_v"):
-      np.testing.assert_equal(getattr(tp, name).numpy(), getattr(single, name).numpy())
-
 class TestPairwiseTopk(unittest.TestCase):
   def test_basic_topk(self):
     x = Tensor([[[1.0, 3.0, 2.0, 5.0, 4.0]]])
@@ -301,13 +213,5 @@ class TestPairwiseTopk(unittest.TestCase):
         self.assertEqual(set(sel.numpy()[b, t].tolist()), expected)
         np.testing.assert_allclose(vals.numpy()[b, t], data[b, t][sel.numpy()[b, t]])
 
-  def test_iterative_matches_numpy(self):
-    rng = np.random.default_rng(42)
-    data = rng.standard_normal((2, 3, 896), dtype=np.float32)
-    vals, sel = iterative_topk(Tensor(data), 16)
-    expected = np.argsort(-data, axis=-1, stable=True)[..., :16]
-    np.testing.assert_equal(sel.numpy(), expected)
-    np.testing.assert_allclose(vals.numpy(), np.take_along_axis(data, expected, axis=-1))
-
 if __name__ == '__main__':
   unittest.main()

test/unit/test_llm_cli_k3.py

@@ -1,30 +0,0 @@
-import unittest
-from tinygrad.llm.cli import KimiK3Template
-from tinygrad.llm.serve import StreamRouter
-
-class TestKimiK3Template(unittest.TestCase):
-  def test_simple_text_chat(self):
-    template = KimiK3Template()
-    got = template.render([{"role":"system", "content":"Be concise."}, {"role":"user", "content":"Hello"}])
-    self.assertTrue(got.startswith('<|open|>message role="system" type="thinking-effort"<|sep|>'))
-    self.assertIn('<|open|>message role="user"<|sep|>Hello<|close|>message<|sep|><|end_of_msg|>', got)
-    self.assertTrue(got.endswith('<|open|>message role="assistant"<|sep|><|open|>think<|sep|>'))
-
-  def test_preserves_assistant_thinking(self):
-    got = KimiK3Template().render([{"role":"assistant", "reasoning_content":"why", "content":"answer"}], add_generation_prompt=False)
-    self.assertIn('<|open|>think<|sep|>why<|close|>think<|sep|>', got)
-    self.assertIn('<|open|>response<|sep|>answer<|close|>response<|sep|>', got)
-
-  def test_rejects_unimplemented_modalities(self):
-    with self.assertRaisesRegex(ValueError, "text-only"):
-      KimiK3Template().render([{"role":"user", "content":[{"type":"image", "url":"x"}]}])
-    with self.assertRaisesRegex(ValueError, "tool rendering"):
-      KimiK3Template().render([{"role":"user", "content":"x"}], tools=[{"type":"function"}])
-
-  def test_xtml_stream_router(self):
-    router, routed = StreamRouter(reasoning=True, xtml=True), []
-    for piece in ("rea", "son<|close|>thi", "nk<|sep|><|open|>response<|sep|>ans", "wer<|close|>response<|sep|>"):
-      routed.extend(router.route(piece))
-    self.assertEqual(routed, [("reasoning_content", "rea"), ("reasoning_content", "son"), ("content", "ans"), ("content", "wer")])
-
-if __name__ == "__main__": unittest.main()

test/unit/test_llm_k3.py

@@ -1,139 +0,0 @@
-import tempfile, unittest
-from pathlib import Path
-from dataclasses import replace
-import numpy as np
-from tinygrad import Tensor, dtypes, nn
-from tinygrad.helpers import getenv
-from tinygrad.llm.kernels import bf16_mfma_splitk
-from tinygrad.llm.kimi_k3 import KIMI_K3_FULL_ATTN_LAYERS, KIMI_K3_SSM_LAYERS, KIMI_K3_TEXT_SIZE, KIMI_K3_TP8_BYTES_PER_GPU, \
-  _layer_sources, _load_stacked_experts, _replace, _safe_load_selected, _shard_kimi_k3, _validate_config, kimi_k3_config, kimi_k3_smoke_config
-from tinygrad.llm.model import FFNBlock, Transformer
-
-def small_k3_config(max_context:int=4): return replace(kimi_k3_smoke_config(max_context), num_experts=8)
-
-class TestKimiK3(unittest.TestCase):
-  def test_selective_safetensor_load(self):
-    with tempfile.TemporaryDirectory() as tmp:
-      path = Path(tmp) / "weights.safetensors"
-      nn.state.safe_save({"keep":Tensor.arange(8), "skip":Tensor.arange(16)}, str(path))
-      selected = _safe_load_selected(path, ["keep"])
-      self.assertEqual(list(selected), ["keep"])
-      np.testing.assert_equal(selected["keep"].numpy(), np.arange(8))
-      with self.assertRaisesRegex(ValueError, "missing tensor absent"): _safe_load_selected(path, ["absent"])
-
-  def test_smoke_config_preserves_gfx950_expert_alignment(self):
-    c = kimi_k3_smoke_config()
-    self.assertEqual(c.routed_expert_dim % 64, 0)
-    self.assertEqual((c.hidden_dim // 8) % 64, 0)
-
-  @unittest.skipUnless(getenv("DEV", "") == "NULL:HIP:gfx950", "gfx950 compile coverage")
-  def test_gfx950_mfma_splitk_compile(self):
-    x = Tensor.zeros(1, 1, 256, dtype=dtypes.bfloat16, device="NULL:HIP:gfx950")
-    weight = Tensor.zeros(16, 256, dtype=dtypes.bfloat16, device="NULL:HIP:gfx950")
-    self.assertEqual(bf16_mfma_splitk(x, weight).realize().shape, (1, 1, 16))
-
-  def test_official_config(self):
-    c = kimi_k3_config(1_048_576)
-    self.assertEqual((c.num_blocks, c.dim, c.n_heads, c.num_experts, c.num_experts_per_tok), (93, 7168, 96, 896, 16))
-    self.assertEqual((sum(KIMI_K3_SSM_LAYERS), len(KIMI_K3_FULL_ATTN_LAYERS)), (69, 24))
-    self.assertEqual(KIMI_K3_FULL_ATTN_LAYERS, (*range(3, 93, 4), 92))
-    self.assertEqual((c.routed_expert_dim, c.hidden_dim, c.shared_expert_dim), (3584, 3072, 6144))
-    self.assertTrue(c.route_weights_uncorrected and c.kda_full_rank_gate and c.attn_output_gate)
-    self.assertTrue(c.ssm is not None and c.ssm.channel_decay)
-    self.assertEqual((c.activation_situ_beta, c.activation_situ_linear_beta, c.kda_gate_lower_bound), (4.0, 25.0, -5.0))
-
-  def test_config_rejects_wrong_checkpoint(self):
-    with self.assertRaisesRegex(ValueError, "not the supported official"):
-      _validate_config({"model_type":"kimi_linear", "hidden_size":2304})
-
-  def test_official_mapping_covers_model(self):
-    model = Transformer(kimi_k3_config(1))
-    state = nn.state.get_state_dict(model)
-    targets = {"token_embd.weight", "output_norm.weight", "output.weight", "output_attn_res_norm.weight", "output_attn_res_proj.weight"}
-    for i,is_kda in enumerate(KIMI_K3_SSM_LAYERS):
-      for target in _layer_sources(i, is_kda).values(): targets.update(target.split("|"))
-      if i:
-        for name in ("ffn_gate_exps.weight", "ffn_gate_exps.weight_scale", "ffn_up_exps.weight", "ffn_up_exps.weight_scale",
-                     "ffn_down_exps.weight", "ffn_down_exps.weight_scale"): targets.add(f"blk.{i}.{name}")
-    self.assertEqual(targets, set(state))
-    self.assertEqual(state["blk.1.ffn_gate_exps.weight"].shape, (896, 3072, 1792))
-    self.assertEqual(state["blk.1.ffn_gate_exps.weight_scale"].shape, (896, 3072, 112))
-    self.assertEqual(state["blk.0.ssm_a"].shape, (128, 1))
-    _shard_kimi_k3(model, tuple(f"NULL:{i}" for i in range(8)))
-    total, per_gpu = 0, 0
-    for name,value in state.items():
-      dtype = dtypes.uint8 if name.endswith(("weight_scale", "_exps.weight")) else dtypes.float32 if name.endswith(
-        ("exp_probs_b.bias", "ssm_q_conv1d.weight", "ssm_k_conv1d.weight", "ssm_v_conv1d.weight", "ssm_norm.weight", "ssm_a", "ssm_dt.bias")) \
-        else dtypes.bfloat16
-      size = value.numel() * dtype.itemsize
-      total += size
-      per_gpu += size if value.uop.axis is None else size//8
-    self.assertEqual((total, per_gpu), (KIMI_K3_TEXT_SIZE, KIMI_K3_TP8_BYTES_PER_GPU))
-
-  def test_situ_matches_reference(self):
-    block = FFNBlock(small_k3_config())
-    gate, up = Tensor([[-8., -1., 0., 3.]]), Tensor([[-30., -2., 5., 40.]])
-    got = block._activation(gate, up).numpy()
-    g, u = gate.numpy().astype(np.float32), up.numpy().astype(np.float32)
-    expected = (4*np.tanh(g/4)/(1+np.exp(-g))) * (25*np.tanh(u/25))
-    np.testing.assert_allclose(got, expected, rtol=1e-5, atol=1e-5)
-
-  def test_attention_residual_matches_reference(self):
-    block = FFNBlock(small_k3_config())
-    block.attn_res_norm.weight.assign([1.0+i/16 for i in range(32)])
-    block.attn_res_proj.weight.assign([[(-1.0)**i/8 for i in range(32)]])
-    prefix, residual = Tensor.arange(64).reshape(2, 32).float()/16, Tensor.arange(128).reshape(2, 2, 32).float()/32
-    got = block._apply_attn_res(prefix, residual, block.attn_res_proj, block.attn_res_norm).numpy()
-    v = np.concatenate((residual.numpy(), prefix.numpy()[:, None]), axis=1).astype(np.float32)
-    k = v / np.sqrt(np.mean(v*v, axis=-1, keepdims=True) + 1e-5)
-    scores = np.sum(k * block.attn_res_norm.weight.numpy() * block.attn_res_proj.weight.numpy()[0], axis=-1)
-    probs = np.exp(scores-scores.max(axis=-1, keepdims=True))
-    probs /= probs.sum(axis=-1, keepdims=True)
-    expected = np.matmul(probs[:, None], v).squeeze(1)
-    np.testing.assert_allclose(got, expected, rtol=1e-5, atol=1e-5)
-
-  def test_tp8_schema(self):
-    model = Transformer(small_k3_config())
-    _shard_kimi_k3(model, tuple(f"NULL:{i}" for i in range(8)))
-    state = nn.state.get_state_dict(model)
-    for name,axis in (("token_embd.weight",0), ("blk.1.ffn_gate_exps.weight",1), ("blk.1.ffn_down_exps.weight_scale",2),
-                      ("blk.1.ffn_routed_down.weight",1), ("blk.0.ssm_g_full.weight",0), ("blk.1.attn_q_b.weight",0)):
-      self.assertEqual(state[name].uop.axis, axis, name)
-    self.assertIsNone(state["blk.1.attn_res_norm.weight"].uop.axis)
-    self.assertIsNone(state["blk.1.ffn_routed_norm.weight"].uop.axis)
-    self.assertIsNone(state["blk.0.ssm_a"].uop.axis)
-
-  def test_direct_expert_staging(self):
-    devices = tuple(f"PYTHON:{i}" for i in range(4))
-    sources = [Tensor([[(e*40+r*4+c)&255 for c in range(4)] for r in range(8)], dtype=dtypes.uint8,
-                      device=devices[0]).realize() for e in range(8)]
-    expected = Tensor.stack(*sources).numpy()
-    for axis in (1, 2):
-      dst = Tensor.zeros(8, 8, 4, dtype=dtypes.uint8, device=devices[0]).shard(devices, axis=axis)
-      _load_stacked_experts(dst, sources)
-      np.testing.assert_equal(dst.numpy(), expected)
-
-  def test_direct_tp_replacement(self):
-    devices = tuple(f"PYTHON:{i}" for i in range(4))
-    source = Tensor.arange(64, dtype=dtypes.float32).reshape(8, 8).realize()
-    expected = source.numpy()
-    for axis in (None, 0, 1):
-      dst = Tensor.zeros(8, 8, device="PYTHON").shard(devices, axis=axis)
-      _replace(dst, source)
-      np.testing.assert_equal(dst.numpy(), expected)
-
-  def test_chunked_recurrent_generate(self):
-    model = Transformer(small_k3_config(max_context=8))
-    for name,value in nn.state.get_state_dict(model).items():
-      fill = 127 if name.endswith("weight_scale") else 0
-      value.replace(Tensor.full(value.shape, fill, dtype=value.dtype if value.dtype is dtypes.uint8 else dtypes.bfloat16, device="PYTHON"))
-    self.assertIsInstance(next(model.generate([1], chunk_size=2)), int)
-    prompt = [1, 2, 3, 4]
-    for _ in range(3): self.assertIsInstance(next(model.generate(prompt.copy(), chunk_size=2)), int)
-    self.assertEqual(model.get_start_pos(model._cached_tokens + [42]), len(prompt))
-    self.assertEqual(model.get_start_pos([9, 2, 3, 4, 42]), 0)
-    self.assertIsInstance(next(model.generate([1, 2, 3, 4, 5], chunk_size=3)), int)
-    self.assertEqual(set(model.recurrent_greedy_prefill_jits), {2})
-    self.assertEqual(model._cached_tokens[:4], [1, 2, 3, 4])
-
-if __name__ == "__main__": unittest.main()

test/unit/test_llm_kimi.py

@@ -1,37 +0,0 @@
-import unittest
-from tinygrad import dtypes, nn
-from tinygrad.llm.kimi import KIMI_LOGICAL_BYTES, KIMI_SSM_LAYERS, KIMI_TENSOR_COUNT, _shard_kimi, _validate_kimi_state, kimi_config
-from tinygrad.llm.model import Transformer
-
-class TestKimiLinear(unittest.TestCase):
-  def test_architecture_config(self):
-    config = kimi_config(4096)
-    self.assertEqual((config.num_blocks, config.dim, config.n_heads, config.vocab_size), (27, 2304, 32, 163840))
-    self.assertEqual(tuple(i for i, is_kda in enumerate(KIMI_SSM_LAYERS) if not is_kda), (3, 7, 11, 15, 19, 23, 26))
-    self.assertEqual((config.num_experts, config.num_experts_per_tok, config.shared_expert_dim), (256, 8, 1024))
-    self.assertTrue(config.expert_mxfp4 and config.bf16_activations and config.kda_split_qkv)
-    self.assertFalse(config.shared_expert_gate)
-
-  def test_tp4_schema_and_axes(self):
-    model = Transformer(kimi_config(32))
-    state = nn.state.get_state_dict(model)
-    self.assertEqual(len(state), KIMI_TENSOR_COUNT)
-    self.assertNotIn("blk.1.ffn_gate_inp_shexp.weight", state)
-    self.assertEqual(state["blk.1.ffn_gate_exps.weight"].dtype, dtypes.uint8)
-    self.assertEqual(state["blk.1.ffn_gate_exps.weight_scale"].dtype, dtypes.uint8)
-
-    _shard_kimi(model, ("NULL:0", "NULL:1", "NULL:2", "NULL:3"))
-    state = nn.state.get_state_dict(model)
-    for name, axis in (("token_embd.weight", 0), ("blk.1.ffn_gate_exps.weight", 1),
-                       ("blk.1.ffn_down_exps.weight_scale", 2), ("blk.3.attn_k_b.weight", 0)):
-      self.assertEqual(state[name].uop.axis, axis, name)
-    self.assertIsNone(state["blk.1.attn_norm.weight"].uop.axis)
-
-  def test_converted_schema_validation(self):
-    model = Transformer(kimi_config(1))
-    state = {name:value if value.dtype is dtypes.uint8 else value.cast(dtypes.bfloat16)
-             for name,value in nn.state.get_state_dict(model).items()}
-    _validate_kimi_state(model, state)
-    self.assertEqual(sum(value.nbytes() for value in state.values()), KIMI_LOGICAL_BYTES)
-
-if __name__ == "__main__": unittest.main()

test/unit/test_llm_moe.py

@@ -72,20 +72,6 @@ class TestMoEFeedForward(unittest.TestCase):
     expected = (Tensor([1.0]).silu().item() + Tensor([3.0]).silu().item()) / 2
     np.testing.assert_allclose(out.numpy()[0, 0, 0], expected, rtol=1e-2)
 
-  def test_kimi_correction_bias_affects_route_weights(self):
-    dim, hidden, n_heads, num_experts, k = 8, 16, 2, 4, 2
-    config = replace(_moe_config(dim, hidden, n_heads, num_experts, k), norm_topk_prob=True, expert_bias=True)
-    block = TransformerBlock(config)
-    block.ffn_gate_exps.weight = Tensor.stack(*[Tensor.eye(hidden, dim) * (i + 1) for i in range(num_experts)])
-    block.ffn_up_exps.weight = Tensor.stack(*[Tensor.eye(hidden, dim) for _ in range(num_experts)])
-    block.ffn_down_exps.weight = Tensor.stack(*[Tensor.eye(dim, hidden) for _ in range(num_experts)])
-    block.ffn_gate_inp.weight = Tensor.zeros(num_experts, dim)
-    block.exp_probs_b["bias"] = Tensor([0.2, 0.1, 0.0, -0.1])
-
-    out = block._feed_forward(Tensor.ones(1, 1, dim))
-    expected = (Tensor([1.0]).silu().item() * 0.7 + Tensor([2.0]).silu().item() * 0.6) / 1.3
-    np.testing.assert_allclose(out.numpy()[0, 0, 0], expected, rtol=1e-2)
-
   def test_moe_feed_forward_shared_expert(self):
     dim, hidden, n_heads = 8, 16, 2
     num_experts, k = 4, 2

test/unit/test_llm_quant.py

@@ -1,68 +0,0 @@
-import unittest
-import numpy as np
-from tinygrad import Tensor, dtypes
-from tinygrad.llm.quant import MXFP4_VALUES, dequantize_mxfp4, quantize_dequantize_mxfp8, quantize_mxfp4, quantize_mxfp4_cpu
-
-class TestMXFormats(unittest.TestCase):
-  def test_mxfp4_known_codes_and_scale(self):
-    values = np.array(MXFP4_VALUES * 2, dtype=np.float32)
-    packed, scale = quantize_mxfp4(Tensor(values))
-    # Positive and negative zero are numerically identical, so nearest-value encoding canonicalizes to +0.
-    np.testing.assert_array_equal(packed.numpy(), np.array([0x10, 0x32, 0x54, 0x76, 0x90, 0xba, 0xdc, 0xfe] * 2, dtype=np.uint8))
-    np.testing.assert_array_equal(scale.numpy(), np.array([127], dtype=np.uint8))
-    np.testing.assert_array_equal(dequantize_mxfp4(packed, scale, dtypes.float32).numpy(), values)
-
-  def test_mxfp4_block_scales_and_zero(self):
-    x = Tensor(np.array([0.0]*32 + [12.0, -12.0] + [0.0]*30, dtype=np.float32))
-    packed, scale = quantize_mxfp4(x)
-    np.testing.assert_array_equal(scale.numpy(), np.array([127, 128], dtype=np.uint8))
-    np.testing.assert_allclose(dequantize_mxfp4(packed, scale, dtypes.float32).numpy(), x.numpy())
-
-  def test_mxfp4_scale_rounds_amax_over_format_max(self):
-    # OCP E8M0 scale selection rounds log2(amax / 6), rather than flooring the
-    # input exponent. At this boundary the two rules differ by a factor of two.
-    x = Tensor(np.array([8.0] + [0.0]*31, dtype=np.float32))
-    packed, scale = quantize_mxfp4(x)
-    np.testing.assert_array_equal(scale.numpy(), np.array([127], dtype=np.uint8))
-    self.assertEqual(dequantize_mxfp4(packed, scale, dtypes.float32).numpy()[0], 6.0)
-
-  def test_mxfp4_cpu_converter_matches_tensor_path(self):
-    x = Tensor(np.linspace(-13, 13, 64*32, dtype=np.float32).reshape(64, 32))
-    packed, scale = quantize_mxfp4(x)
-    cpu_packed, cpu_scale = quantize_mxfp4_cpu(x)
-    np.testing.assert_array_equal(cpu_packed.numpy(), packed.numpy())
-    np.testing.assert_array_equal(cpu_scale.numpy(), scale.numpy())
-
-  def test_mxfp4_midpoints_round_to_even(self):
-    midpoints = np.array([0.25, 0.75, 1.25, 1.75, 2.5, 3.5, 5.0], dtype=np.float32)
-    x = Tensor(np.pad(np.concatenate((midpoints, -midpoints)), (0, 18)))
-    packed, scale = quantize_mxfp4(x)
-    expected = np.pad(np.array([0, 1, 1, 2, 2, 4, 4, 0, -1, -1, -2, -2, -4, -4], dtype=np.float32), (0, 18))
-    np.testing.assert_array_equal(dequantize_mxfp4(packed, scale, dtypes.float32).numpy(), expected)
-
-  def test_mxfp8_roundtrip_and_dtype(self):
-    # All E4M3-exact values remain exact after extracting a shared exponent.
-    x = Tensor(np.array(([0.0, 0.5, 1.0, 1.5, 2.0, -3.0, 4.0, -6.0] * 4), dtype=np.float32))
-    out = quantize_dequantize_mxfp8(x)
-    self.assertEqual(out.dtype, dtypes.bfloat16)
-    np.testing.assert_array_equal(out.float().numpy(), x.numpy())
-
-  def test_mxfp8_subnormal_and_rounding(self):
-    x = np.zeros(32, dtype=np.float32)
-    x[:5] = [1.0, 1.0625, 1.07, 2**-9, 2**-10]
-    out = quantize_dequantize_mxfp8(Tensor(x), dtype=dtypes.float32).numpy()
-    # amax / 448 rounds to an E8M0 scale of 2**-9, saturating the largest
-    # values while retaining the E4M3 subnormal quantum for this block.
-    np.testing.assert_array_equal(out[:5], [0.875, 0.875, 0.875, 2**-9, 2**-10])
-
-  def test_mxfp8_uses_full_e4m3_range(self):
-    x = np.zeros(32, dtype=np.float32)
-    x[:4] = [448.0, 416.0, 400.0, -448.0]
-    np.testing.assert_array_equal(quantize_dequantize_mxfp8(Tensor(x), dtype=dtypes.float32).numpy()[:4], [448.0, 416.0, 384.0, -448.0])
-
-  def test_mxfp8_scale_rounds_amax_over_format_max(self):
-    x = np.zeros(32, dtype=np.float32)
-    x[0] = 512.0
-    self.assertEqual(quantize_dequantize_mxfp8(Tensor(x), dtype=dtypes.float32).numpy()[0], 448.0)
-
-if __name__ == "__main__": unittest.main()

test/unit/test_llm_server.py

@@ -1,10 +1,9 @@
 import unittest
-from dataclasses import replace
 from unittest.mock import patch
 from tinygrad import Tensor, UOp
 from tinygrad.schedule import schedule_cache
 from tinygrad.llm.model import Transformer, TransformerConfig
-from tinygrad.llm.serve import StreamRouter, parse_kimi_tool_call
+from tinygrad.llm.serve import StreamRouter
 
 TEST_CONFIG = TransformerConfig(num_blocks=1, dim=64, hidden_dim=128, n_heads=2, n_kv_heads=2,
                            norm_eps=1e-5, vocab_size=100, head_dim=32, rope_theta=10000.0, rope_dim=32, v_head_dim=32, max_context=32)
@@ -14,29 +13,11 @@ V_TOKS = UOp.variable("toks", 1, 32)  # 32 is the default chunk_size in generate
 class TestTransformerGenerate(unittest.TestCase):
   def test_warmup(self):
     model, calls = Transformer(TEST_CONFIG), []
-    def generate(tokens, temperature):
-      calls.append((tokens, temperature))
+    def generate(tokens):
+      calls.append(tokens)
       yield from (1, 2)
     with patch.object(model, "generate", generate): model.warmup()
-    self.assertEqual(calls, [([0], 0.0), ([0], 0.0)])
-
-  def test_recurrent_warmup_captures_reset_replay(self):
-    model, calls = Transformer(TEST_CONFIG), []
-    model.has_recurrent_block = True
-    state = Tensor.ones(4).realize()
-    model.blk[0]._state_reset_ops = lambda: [state.assign(state.const_like(0))]
-    def generate(tokens, temperature):
-      if calls: model.reset_jit()
-      calls.append((tokens.copy(), temperature))
-      tokens.append(42)
-      yield from (1, 2)
-    with patch.object(model, "generate", generate): model.warmup()
-    prompt = [0] * (TEST_CONFIG.max_context-2)
-    self.assertEqual(calls, [(prompt, 0.0)] * 3 + [(prompt, 1.0)] * 3 + [(prompt + list(range(1, i+1)), 0.0) for i in range(1, 4)])
-    self.assertEqual(model.reset_jit.cnt, 8)
-    cache_size = len(schedule_cache)
-    model.reset_jit()
-    self.assertEqual(len(schedule_cache), cache_size)
+    self.assertEqual(calls, [[0], [0]])
 
   def test_first_recurrent_generate_before_state_init(self):
     model = Transformer(TEST_CONFIG)
@@ -44,17 +25,6 @@ class TestTransformerGenerate(unittest.TestCase):
     with patch.object(Transformer, '__call__', return_value=Tensor([[42]])):
       self.assertEqual(next(model.generate([0])), 42)
 
-  def test_recurrent_prefill_tail_uses_rollout_shape(self):
-    model = Transformer(TEST_CONFIG)
-    model.has_recurrent_block = True
-    model.config = replace(model.config, recurrent_prefill_chunked=True)
-    calls = []
-    def mock_call(self, tokens, start_pos, temperature, **kwargs):
-      calls.append(tokens.shape)
-      return Tensor([[42]])
-    with patch.object(Transformer, '__call__', mock_call): next(model.generate([1, 2, 3, 4, 5, 6], chunk_size=4))
-    self.assertEqual(calls, [(1, 4), (1, 1), (1, 1)])
-
   def test_recurrent_live_state_reuse(self):
     model = Transformer(TEST_CONFIG)
     model.has_recurrent_block = True
@@ -68,37 +38,11 @@ class TestTransformerGenerate(unittest.TestCase):
       next(model.generate([1, 2, 3, 4, 5, 42, 10]))
     self.assertEqual(calls, [((1, 1), V_START_POS.bind(5)), ((1, 1), V_START_POS.bind(6))])
 
-  def test_recurrent_prompt_snapshot_reuse(self):
-    model = Transformer(TEST_CONFIG)
-    model.has_recurrent_block = True
-    state, calls = Tensor.ones(4).realize(), []
-    def mock_call(self, tokens, start_pos, temperature, **kwargs):
-      calls.append(start_pos)
-      return Tensor([[42]])
-    with patch.object(model, "_state_tensors", return_value=[state]), patch.object(model.blk[0], "_reusable_prefix_len", return_value=0), \
-         patch.object(Transformer, '__call__', mock_call):
-      next(model.generate([1, 2, 3]))
-      state.assign(state.const_like(5)).realize()
-      model._cached_tokens = [1, 2, 3, 9, 9]
-      calls.clear()
-      self.assertEqual(model.get_start_pos([1, 2, 3, 7, 8]), 3)
-      next(model.generate([1, 2, 3, 7, 8]))
-    self.assertEqual(calls, [V_START_POS.bind(3), V_START_POS.bind(4)])
-    self.assertEqual(state.tolist(), [1.0] * 4)
-
   def test_template_starts_reasoning(self):
     router = StreamRouter(reasoning=True)
     self.assertEqual(list(router.route("reasoning</think>answer")),
                      [("reasoning_content", "reasoning"), ("content", "answer")])
 
-  def test_kimi_tool_call_stream(self):
-    router = StreamRouter()
-    self.assertEqual(list(router.route("before<|tool_calls_section_beg")), [("content", "before")])
-    self.assertEqual(list(router.route("in|><|tool_call_begin|>functions.read:0<|tool_call_argument_begin|>"
-                                      '{"path":"/tmp/x"}<|tool_call_end|><|tool_calls_section_end|>')), [])
-    self.assertEqual(parse_kimi_tool_call("functions.read:0<|tool_call_argument_begin|>{\"path\":\"/tmp/x\"}"),
-                     ("read", {"path":"/tmp/x"}))
-
   def test_kv_cache_reuse(self):
     """Test that generate reuses the KV cache when tokens extend the cached prefix."""
     model = Transformer(TEST_CONFIG)

test/unit/test_metal_graph.py

@@ -1,5 +1,4 @@
 import unittest
-from unittest.mock import MagicMock
 from tinygrad import Device
 from tinygrad.uop.ops import Ops, UOp
 from tinygrad.dtype import dtypes
@@ -11,36 +10,27 @@ class TestMetalGraph(unittest.TestCase):
     self.MetalGraph = MetalGraph
     self.dev = Device[Device.DEFAULT]
 
-  def metal_buf(self, offset):
-    buf = MagicMock()
-    if offset > 0:
-      buf.op = Ops.SLICE
-      src = MagicMock()
-      src.dtype = dtypes.uint8
-      buf.src = (src, UOp.const(offset))
-      buf.dtype = dtypes.uint8
-    else:
-      buf.op = Ops.BUFFER
-    buf.device = Device.DEFAULT
-    return buf
+  def metal_buf(self, offset, bitcast=False):
+    size = 4 if bitcast else 1
+    buf = UOp.new_buffer(Device.DEFAULT, offset+size, dtypes.uint8)
+    if offset: buf = buf[offset:offset+size]
+    return buf.bitcast(dtypes.float32) if bitcast else buf
 
-  def call(self, *bufs):
-    c = MagicMock()
-    c.src = (MagicMock(op=Ops.PROGRAM),) + tuple(bufs)
-    return c
+  def supports_uop(self, *bufs):
+    return self.MetalGraph.supports_uop([self.dev], UOp(Ops.PROGRAM, src=(UOp.sink(),)).call(*bufs))
 
   def test_supports_uop_normal_offset(self):
-    assert self.MetalGraph.supports_uop([self.dev], self.call(self.metal_buf(0), self.metal_buf(100), self.metal_buf(0xFFFFFFFF))) is True
+    assert self.supports_uop(self.metal_buf(0), self.metal_buf(100), self.metal_buf(0xFFFFFFFF)) is True
 
   def test_supports_uop_overflow_offset(self):
-    assert self.MetalGraph.supports_uop([self.dev], self.call(self.metal_buf(0), self.metal_buf(0x100000000))) is False
+    assert self.supports_uop(self.metal_buf(0), self.metal_buf(0x100000000)) is False
 
-  def test_supports_uop_nonmetal_buf(self):
-    # non-SLICE ops should not be checked for offset
-    buf = MagicMock()
-    buf.op = Ops.BUFFER
-    buf.device = Device.DEFAULT
-    self.MetalGraph.supports_uop([self.dev], self.call(buf))
+  def test_supports_uop_non_view_buf(self):
+    assert self.supports_uop(self.metal_buf(0)) is True
+
+  def test_supports_uop_bitcast(self):
+    assert self.supports_uop(self.metal_buf(0xFFFFFFFF, bitcast=True)) is True
+    assert self.supports_uop(self.metal_buf(0x100000000, bitcast=True)) is False
 
 if __name__ == "__main__":
   unittest.main()

tinygrad/codegen/opt/tc.py

@@ -1,6 +1,7 @@
 import math, functools
 from dataclasses import dataclass
 from tinygrad.dtype import DType, dtypes
+from tinygrad.uop.ops import PatternMatcher, UOp, UPat, Ops
 
 @dataclass(frozen=True)
 class TensorCore: # D = A * B + C, A is (M x K), B is (K x N), C and D are (M x N)
@@ -135,6 +136,42 @@ amd_cdna4 = amd_cdna_1616128 + amd_cdna_161632 + amd_cdna_161616
 
 def get_amd(arch): return {"gfx942": amd_cdna3, "gfx950": amd_cdna4, "gfx1200": amd_rdna4, "gfx1201": amd_rdna4}.get(arch, amd_rdna3)
 
+pm_validate_wmma_rdna3 = PatternMatcher([
+  (UPat(Ops.WMMA, name="x", dtype=dtypes.int32), lambda x: x.replace(
+    src=(x.src[0].bitcast(dtypes.uint32), x.src[1].bitcast(dtypes.uint32), x.src[2]))
+    if x.src[0].dtype == dtypes.int8 and x.src[0].max_numel() == 16 else None),
+  (UPat(Ops.WMMA, name="x", dtype=dtypes.half), lambda x: UOp(Ops.STACK, src=tuple(x.replace(
+      src=(x.src[0], x.src[1], UOp(Ops.STACK, src=tuple(x.src[2].index(UOp.const(j//2, dtypes.int16))
+      if j%2 == 0 else UOp.const(0.0, x.src[2].dtype)
+      for j in range(x.max_numel()*2)))),
+      arg=(*x.arg[:4], None)).index(UOp.const(i*2, dtypes.int16))
+      for i in range(x.max_numel()))) if x.max_numel() == 8 else None),
+  (UPat(Ops.WMMA, name="x"), lambda x: x.replace(
+    src=(x.src[0].bitcast(dtypes.uint16), x.src[1].bitcast(dtypes.uint16), x.src[2]))
+    if x.src[0].dtype == dtypes.bfloat16 and x.src[0].max_numel() == 16 else None),
+])
+
+pm_validate_wmma_rdna4 = PatternMatcher([
+  (UPat(Ops.WMMA, name="x", dtype=dtypes.bfloat16), lambda x: x.replace(
+    dtype=dtypes.uint16,
+    src=(x.src[0].bitcast(dtypes.uint16), x.src[1].bitcast(dtypes.uint16), x.src[2].bitcast(dtypes.uint16)))
+      .bitcast(dtypes.bfloat16) if x.max_numel() == 8 and x.src[0].dtype == dtypes.bfloat16 and x.src[0].max_numel() == 8 else None),
+  (UPat(Ops.WMMA, name="x", dtype=dtypes.float),
+    lambda x: x.replace(src=(x.src[0].bitcast(dtypes.uint16), x.src[1].bitcast(dtypes.uint16), x.src[2]))
+    if x.max_numel() == 8 and x.src[0].dtype == dtypes.bfloat16 and x.src[0].max_numel() == 8 else None)
+])
+
+pm_validate_wmma_cdna = PatternMatcher([
+  (UPat(Ops.WMMA, name="x", dtype=dtypes.float),
+    lambda x: x.replace(src=(x.src[0].bitcast(dtypes.uint32), x.src[1].bitcast(dtypes.uint32), x.src[2]))
+    if x.arg[0][2] == 128 and x.src[0].dtype.itemsize <= 8 else None),
+  (UPat(Ops.WMMA, name="x", dtype=dtypes.float),
+    lambda x: x.replace(src=(x.src[0].bitcast(dtypes.uint16), x.src[1].bitcast(dtypes.uint16), x.src[2]))
+    if x.max_numel() == 4 and x.src[0].dtype == dtypes.bfloat16 and x.src[0].max_numel() == 4 else None),
+  (UPat(Ops.WMMA, name="x", dtype=dtypes.float),
+    lambda x: x.replace(src=(x.src[0].bitcast(dtypes.uint64), x.src[1].bitcast(dtypes.uint64), x.src[2]))
+    if x.max_numel() == 4 and x.src[0].dtype in dtypes.fp8_ocp and x.src[0].max_numel() == 8 else None),
+])
 # ***** Apple Metal *****
 
 metal = [TensorCore(dims=(8,8,8), threads=32, elements_per_thread=(2,2,2), dtype_in=di, dtype_out=do,

tinygrad/device.py

@@ -336,6 +336,8 @@ class Compiled:
   pm_lower:Any = None
   pm_bufferize:Any = None
 
+  has_copy_queue:bool = True
+
   def __init__(self, device:str, allocator:Allocator, renderers:list[type[Renderer]], runtime:type[Program[Self]]|None, graph=None, arch=None):
     from tinygrad.renderer import Renderer
     self.device, self.allocator, self.runtime_t, self.graph, self.renderers = device, allocator, runtime, graph, renderers or [Renderer]

tinygrad/engine/jit.py

@@ -44,8 +44,6 @@ def graph_split_rewrite(linear:UOp, max_batch_size:int=0) -> UOp:
     current_batch, current_batch_devs = [], []
 
   for si in linear.src:
-    if si.src[0].op is Ops.SLICE: continue
-
     devs = dedup([Device[x] for b in si.src[1:] if b.op is not Ops.BIND for x in (b.device if isinstance(b.device, tuple) else (b.device,))])
     graph_t = graph_class(devs[0]) if devs[0].graph is not None else None
 
@@ -180,7 +178,7 @@ class CapturedJit(Generic[ReturnType]):
       if call.op is not Ops.CALL: continue
       arg_uops = get_call_arg_uops(call)
       outs, ins = get_call_outs_ins(call)
-      out |= {arg_uops[k] for k in set(outs) - set(ins) if arg_uops[k].op in (Ops.BUFFER, Ops.SLICE)}
+      out |= {b for k in set(outs) - set(ins) if (b:=u if (cv:=(u:=arg_uops[k]).contiguous_view()) is None else cv[0]).op is Ops.BUFFER}
     return out
 
   def __call__(self, input_uops:list[UOp], var_vals:dict[str, int]) -> ReturnType:

tinygrad/engine/realize.py

@@ -4,8 +4,8 @@ import time, random, itertools, math, contextlib, weakref, array
 from dataclasses import dataclass, replace, field
 from tinygrad.helpers import colored, DEBUG, GlobalCounters, ansilen, all_int, prod, flatten, Context, getenv, to_tuple
 from tinygrad.helpers import BEAM, size_to_str, time_to_str, VALIDATE_WITH_CPU, PROFILE, ProfilePointEvent, cpu_events, wait_cond
-from tinygrad.uop.ops import Ops, PatternMatcher, UOp, UPat, AxisType, sym_infer, buffers, graph_rewrite
-from tinygrad.device import Device, Buffer, MultiBuffer
+from tinygrad.uop.ops import Ops, PatternMatcher, UOp, UPat, AxisType, sym_infer, graph_rewrite
+from tinygrad.device import Device, Buffer, MultiBuffer, ProfileGraphEntry
 from tinygrad.renderer import Estimates
 from tinygrad.codegen import to_program
 from tinygrad.codegen.opt.postrange import args_from_ast
@@ -17,7 +17,7 @@ def get_call_arg_uops(call:UOp) -> tuple[UOp, ...]: return tuple(s for s in call
 def get_call_outs_ins(call:UOp) -> tuple[tuple[int, ...], tuple[int, ...]]:
   ast = call.src[0]
   if ast.op is Ops.PROGRAM: return tuple(ast.arg.outs), tuple(ast.arg.ins)
-  if ast.op in (Ops.COPY, Ops.SLICE): return (0,), (1,)
+  if ast.op is Ops.COPY: return (0,), (1,)
   if ast.op is Ops.CUSTOM_FUNCTION and ast.arg == "encdec": return (0,), tuple(range(1, len(get_call_arg_uops(call))))
   return (), ()
 
@@ -27,9 +27,6 @@ def get_call_name(call:UOp, bufs:Sequence[Buffer|UOp], var_vals:dict[str, int]|N
 
   ast, arg_uops = call.src[0], get_call_arg_uops(call)
   if ast.op is Ops.PROGRAM: return ast.arg.name
-  if ast.op is Ops.SLICE:
-    offset = ast.src[1].val * arg_uops[1].dtype.itemsize
-    return colored(f"view {_uop_sz_to_str(arg_uops[0]):>10} @ {offset:<10d}", "yellow")
   if ast.op is Ops.COPY: return colored(f"copy {_uop_sz_to_str(arg_uops[0]):>10}, {_dev_str(bufs[0]):>7s} <- {_dev_str(bufs[1]):7s}", "yellow")
   if ast.op is Ops.CUSTOM_FUNCTION and ast.arg == "encdec": return colored(f"enc/dec {_uop_sz_to_str(arg_uops[0])}", "yellow")
   if ast.op is Ops.CUSTOM_FUNCTION and ast.arg == "graph": return colored(f"batched {len(ast.src[0].src)}", "cyan")
@@ -140,7 +137,7 @@ class ExecContext:
   cache: bool = True
 
 def _resolve(b:UOp, inputs:tuple[UOp, ...]) -> UOp:
-  if b.op in (Ops.SLICE, Ops.MSELECT) and b.src[0].op is Ops.PARAM: return b.replace(src=(inputs[b.src[0].arg.slot], *b.src[1:]))
+  if b.op in (Ops.MSELECT, Ops.SHRINK) and b.src[0].op is Ops.PARAM: return b.replace(src=(inputs[b.src[0].arg.slot], *b.src[1:]))
   if b.op is Ops.MSTACK: return b.replace(src=tuple(_resolve(x, inputs) for x in b.src))
   return inputs[b.arg.slot] if b.op is Ops.PARAM else b
 def resolve_params(call:UOp, inputs:tuple[UOp, ...]) -> list[UOp]: return [_resolve(b, inputs) for b in get_call_arg_uops(call)]
@@ -154,13 +151,6 @@ def unwrap_multi(call:UOp, resolved:list[UOp]) -> Iterator[tuple[list[Buffer], d
                    for x in call.src[0].toposort())
     for j, per_dev in enumerate(zip(*[cast(MultiBuffer, b).bufs for b in bufs])): yield list(per_dev), {"_device_num": j} if has_dnum else {}
 
-def exec_view(ctx:ExecContext, call:UOp, ast:UOp) -> float|None:
-  resolved = resolve_params(call, ctx.input_uops)
-  bufs = [cast(Buffer, b.buffer) for b in resolved]
-  bv = bufs[1].view(resolved[0].max_numel(), ast.dtype, ast.src[1].val*bufs[1].dtype.itemsize)
-  with track_stats(ctx, call, bv.device, [bv, bufs[1]], ctx.var_vals): buffers[resolved[0]] = bv
-  return None
-
 def exec_copy(ctx:ExecContext, call:UOp, ast:UOp) -> float|None:
   for bufs, device_vars in unwrap_multi(call, resolve_params(call, ctx.input_uops)):
     dest, src = bufs[0].ensure_allocated(), bufs[1].ensure_allocated()
@@ -210,27 +200,30 @@ def exec_graph(ctx:ExecContext, call:UOp, ast:UOp) -> float|None:
   return t[0]
 
 def exec_hcq(ctx:ExecContext, call:UOp, ast:UOp) -> float|None:
-  if (inputs:=call.arg.aux.inputs) is not None:
+  if (info:=call.arg.aux).inputs is not None:
     bufs = [_resolve(ctx.input_uops[i], ctx.input_uops).buffer for i in call.arg.aux.input_idxs]
-    table = call.src[1+inputs].buffer
+    table = call.src[1+info.inputs].buffer
     for j,dev in enumerate(call.arg.aux.device):
       addrs = array.array('Q', [(b.bufs[j] if isinstance(b, MultiBuffer) else b).get_buf(dev).va_addr for b in bufs])
       mv = (table.bufs[j] if isinstance(table, MultiBuffer) else table).ensure_allocated()._buf.cpu_view().view(fmt='Q')
       wait_cond(lambda: mv[0], value=0, timeout_ms=ctx.timeout or getenv("HCQDEV_WAIT_TIMEOUT_MS", 30000), msg=f"{dev} hang detected")
       mv[:len(addrs)] = addrs
 
-  exec_kernel(replace(ctx, update_stats=False), call, ast)
+  exec_kernel(replace(ctx, update_stats=DEBUG>=3), call, ast)
 
-  tms:list[float|None] = []
-  for e in (aux:=call.arg.aux).prof: cast(Any, Device[e.device]).prof_ents[e.st_id] = e
-  for d in [cast(Any, Device[x]) for x in aux.device]:
-    with track_stats(ctx, call, d.device, [], ctx.var_vals) as et:
-      if ctx.wait:
-        d.synchronize(timeout=ctx.timeout)
-        ts = [d.signal(i)._buf.cpu_view().view(fmt='Q')[0] for e in aux.prof if e.device == d.device for i in (e.st_id, e.en_id)]
-        if ts: et[0] = float(max(ts)-min(ts))/d.timestamp_divider/1e6
-      tms += et
-  return tms[0]
+  tms = []
+  for devices,name,estimates,prof in info.kernels:
+    for device in devices:
+      d, tm = cast(Any, Device[device]), None
+      if prof:
+        d.prof_ents[prof[0]] = ProfileGraphEntry(device, name, *prof)
+        if ctx.wait:
+          d.synchronize(timeout=ctx.timeout)
+          st, en = (d.signal(x)._buf.cpu_view().view(fmt='Q')[0] for x in prof)
+          tms.append(tm:=float(en-st)/d.timestamp_divider/1e6)
+      with track_stats(ctx, call.replace(arg=replace(call.arg, name=name, aux=replace(info, estimates=estimates))), d.device, [], ctx.var_vals) as et:
+        et[0] = tm
+  return max(tms) if tms else None
 
 # flatten LINEAR-in-LINEAR: any nested LINEAR child gets inlined into its parent's src
 pm_flatten_linear = PatternMatcher([
@@ -261,7 +254,6 @@ pm_optimize_local_size = PatternMatcher([
 ])
 
 pm_exec = PatternMatcher([
-  (UPat(Ops.CALL, src=(UPat(Ops.SLICE, name="ast"),), name="call", allow_any_len=True), exec_view),
   (UPat(Ops.CALL, src=(UPat(Ops.COPY, name="ast"),), name="call", allow_any_len=True), exec_copy),
   (UPat(Ops.CALL, src=(UPat(Ops.PROGRAM, name="ast"),), name="call", allow_any_len=True), exec_kernel),
   (UPat(Ops.CALL, src=(UPat(Ops.CUSTOM_FUNCTION, arg="encdec", name="ast"),), name="call", allow_any_len=True), exec_encdec),
@@ -276,7 +268,7 @@ def compile_linear(linear:UOp, beam:int|None=None, validate=False, input_uops:li
   if validate: linear = graph_rewrite(linear, pm_validate, name="validate", walk=True)
   if (beam_val:=BEAM.value if beam is None else beam) >= 1: linear = graph_rewrite(linear, pm_beam, ctx=beam_val, walk=True)
   linear = graph_rewrite(linear, pm_compile, name="precompile kernels", walk=True)
-  if getenv("HCQ2"): linear = hcq_compile(linear, input_uops, bool(PROFILE) if profile is None else profile)
+  if getenv("HCQ2"): linear = hcq_compile(linear, input_uops, bool(PROFILE or DEBUG >= 2) if profile is None else profile)
   return graph_rewrite(linear, pm_optimize_local_size, name="optimize local size", walk=True)
 
 def link_linear(linear:UOp, cache=True) -> UOp: return hcq_link(linear, cache=cache) if getenv("HCQ2") else linear

tinygrad/helpers.py

@@ -1,9 +1,9 @@
 from __future__ import annotations
 import time
 START_TIME = time.perf_counter()
-import os, functools, platform, re, contextlib, operator, hashlib, pickle, sqlite3, tempfile, pathlib, string, ctypes, sys, gzip, getpass, gc
+import os, functools, re, contextlib, operator, hashlib, pickle, sqlite3, tempfile, pathlib, string, ctypes, sys, gzip, getpass, gc
 from collections import defaultdict
-import subprocess, shutil, math, types, copyreg, inspect, importlib, decimal, itertools, difflib
+import shutil, math, types, copyreg, inspect, importlib, decimal, itertools, difflib
 from dataclasses import dataclass, field, replace
 from typing import ClassVar, Iterable, Any, TypeVar, Callable, Sequence, TypeGuard, Iterator, Generic, Generator, cast, overload
 
@@ -13,8 +13,7 @@ U = TypeVar("U")
 def prod(x:Iterable[T]) -> T|int: return functools.reduce(operator.mul, x, 1)
 
 # NOTE: helpers is not allowed to import from anything else in tinygrad
-OSX, WIN = platform.system() == "Darwin", sys.platform == "win32"
-ARCH_X86 = any(x in platform.processor() for x in ("Intel", "i386", "x86_64"))
+OSX, WIN = sys.platform == "darwin", sys.platform == "win32"
 BASEDIR = pathlib.Path(__file__).parent
 
 # fix colors on Windows, https://stackoverflow.com/questions/12492810/python-how-can-i-make-the-ansi-escape-codes-to-work-also-in-windows
@@ -231,7 +230,7 @@ class _DEV(ContextVar):
 
 DEV, DEBUG, BEAM, NOOPT = _DEV("DEV", ""), ContextVar("DEBUG", 0), ContextVar("BEAM", 0), ContextVar("NOOPT", 0)
 IMAGE, FLOAT16, OPENPILOT_HACKS = ContextVar("IMAGE", 0), ContextVar("FLOAT16", 0), ContextVar("OPENPILOT_HACKS", 0)
-JIT, JIT_BATCH_SIZE = ContextVar("JIT", 2 if OSX and ARCH_X86 else 1), ContextVar("JIT_BATCH_SIZE", 32)
+JIT, JIT_BATCH_SIZE = ContextVar("JIT", 1), ContextVar("JIT_BATCH_SIZE", 32)
 CHUNK_SIZE = 2**20  # TinyFS content-addressed store: blob chunk + hash-tree node granularity
 WINO, CAPTURING, TRACEMETA, NO_COLOR = ContextVar("WINO", 0), ContextVar("CAPTURING", 1), ContextVar("TRACEMETA", 1), ContextVar("NO_COLOR", 0)
 TRAINING = ContextVar("TRAINING", 0)
@@ -454,9 +453,9 @@ def _ensure_downloads_dir() -> pathlib.Path:
   if pathlib.Path("/etc/tinybox-release").is_file():
     # try creating dir with sudo
     if not (downloads_dir := pathlib.Path("/raid/downloads")).exists():
-      subprocess.run(["sudo", "mkdir", "-p", downloads_dir], check=True)
-      subprocess.run(["sudo", "chown", "tiny:root", downloads_dir], check=True)
-      subprocess.run(["sudo", "chmod", "775", downloads_dir], check=True)
+      system(f"sudo mkdir -p {downloads_dir}")
+      system(f"sudo chown tiny:root {downloads_dir}")
+      system(f"sudo chmod 775 {downloads_dir}")
     return downloads_dir
   return pathlib.Path(cache_dir) / "downloads"
 
@@ -497,6 +496,7 @@ def fetch_fw(path:str, name:str, sha256:str) -> bytes:
 # *** Exec helpers
 
 def system(cmd:str, **kwargs) -> str:
+  import subprocess
   st = time.perf_counter()
   try: ret = subprocess.check_output(cmd.split(), stderr=subprocess.STDOUT, **kwargs).decode().strip()
   except subprocess.CalledProcessError as e:

tinygrad/llm/cli.py

@@ -1,5 +1,5 @@
 from __future__ import annotations
-import sys, argparse, codecs, itertools, typing, re, unicodedata, json, time, pathlib
+import sys, argparse, codecs, itertools, typing, re, unicodedata, json, time
 from typing import TYPE_CHECKING
 from tinygrad import nn
 from tinygrad.uop.ops import UOp, Ops
@@ -127,95 +127,30 @@ class FallbackTemplate:
       out += self.end_turn()
     return out + self.role("assistant") if add_generation_prompt else out
 
-class KimiK3Template:
-  """Official K3 XTML envelope for text-only system/user/assistant conversations."""
-  OPEN, CLOSE, SEP, END = "<|open|>", "<|close|>", "<|sep|>", "<|end_of_msg|>"
-  def _open(self, tag:str, attrs:tuple[tuple[str, str], ...]=()) -> str:
-    escaped = ((k, str(v).replace("&", "&amp;").replace('"', "&quot;")) for k,v in attrs)
-    return self.OPEN + tag + "".join(f' {k}="{v}"' for k,v in escaped) + self.SEP
-  def _close(self, tag:str) -> str: return self.CLOSE + tag + self.SEP
-  def _message(self, role:str, content:str, name:str|None=None) -> str:
-    attrs = (("role", role),) + (() if name is None else (("name", name),))
-    return self._open("message", attrs) + content + self._close("message") + self.END
-  @staticmethod
-  def _content(message:dict) -> str:
-    content = message.get("content")
-    if content is None: return ""
-    if isinstance(content, str): return content
-    if isinstance(content, list):
-      if any(part.get("type") != "text" for part in content): raise ValueError("Kimi K3 native loader is text-only; image content is not implemented")
-      return "".join(part["text"] for part in content)
-    raise ValueError(f"unsupported Kimi K3 content type {type(content).__name__}")
-  def render(self, messages:list[dict], tools=None, add_generation_prompt:bool=True, preserve_thinking:bool=False, **kwargs) -> str:
-    if tools or any(m.get("role") == "tool" or m.get("tool_calls") for m in messages):
-      raise ValueError("Kimi K3 XTML tool rendering is not implemented in the native text loader")
-    effort = kwargs.get("thinking_effort", "max")
-    if effort not in ("low", "high", "max"): raise ValueError(f"invalid Kimi K3 thinking_effort {effort!r}")
-    body = "`thinking_effort` guides on how much to think in your thinking channel (not including the response channel), " \
-           "supported values include `low`, `medium`, `high`, and `max`.\n" \
-           f"Now the system is invoked with `thinking_effort={effort}`."
-    out = self._open("message", (("role", "system"), ("type", "thinking-effort"))) + body + self._close("message") + self.END
-    for message in messages:
-      role = message["role"]
-      if role in ("user", "system"):
-        out += self._message(role, self._content(message), message.get("name"))
-      elif role == "assistant":
-        reasoning = message.get("reasoning_content") or message.get("reasoning") or ""
-        content = self._open("think") + str(reasoning) + self._close("think")
-        content += self._open("response") + self._content(message) + self._close("response")
-        out += self._message(role, content, message.get("name"))
-      else: raise ValueError(f"unsupported Kimi K3 role {role!r}")
-    if add_generation_prompt: out += self._open("message", (("role", "assistant"),)) + self._open("think")
-    return out
-
-from tinygrad.llm.serve import LLMServer, StreamRouter
+from tinygrad.llm.serve import LLMServer
 
 def main():
   parser = argparse.ArgumentParser()
-  parser.add_argument("--model", "-m", default=list(models.keys())[0],
-                      help=f"Model choice ({', '.join(models.keys())}), local GGUF file, converted Kimi directory, or official Kimi K3 directory")
+  parser.add_argument("--model", "-m", default=list(models.keys())[0], help=f"Model choice ({', '.join(models.keys())}) or path to a local GGUF file")
   parser.add_argument("--max_context", type=int, default=4096, help="Max Context Length")
   parser.add_argument("--serve", nargs='?', type=int, const=8000, metavar="PORT", help="Run OpenAI compatible API (optional port, default 8000)")
   parser.add_argument("--warmup", action="store_true", help="warmup the JIT")
   parser.add_argument("--benchmark", nargs='?', type=int, const=20, metavar="COUNT", help="Benchmark tok/s (optional count, default 20)")
-  parser.add_argument("--devices", type=int, default=1, help="Tensor-parallel device count (Kimi-Linear requires 4, Kimi K3 requires 8)")
   args = parser.parse_args()
 
   # load the model
-  model_path = pathlib.Path(args.model)
-  kv:dict[str, typing.Any]
-  is_k3 = False
-  if model_path.is_dir() and (model_path / "config.json").exists():
-    raw_config = json.loads((model_path / "config.json").read_text())
-    is_k3 = raw_config.get("model_type") == "kimi_k3"
-  if is_k3:
-    from tinygrad.llm.kimi_k3 import load_kimi_k3, load_kimi_tokenizer_data
-    model, kv = load_kimi_k3(model_path, args.max_context, args.devices), {}
-    normal, special, bos, eos = load_kimi_tokenizer_data(model_path)
-    tok = SimpleTokenizer(normal, special, "kimi-k2", bos_id=bos, eos_id=eos, eot_id=eos)
-    model_name = "Kimi-K3"
-    tok_cfg = json.loads((model_path / "tokenizer_config.json").read_text())
-    ct = tok_cfg.get("chat_template")
-  elif model_path.is_dir() and (model_path / "tinygrad-kimi.json").exists():
-    from tinygrad.llm.kimi import load_kimi, load_kimi_tokenizer_data
-    model, kv = load_kimi(model_path, args.max_context, args.devices), {}
-    normal, special, bos, eos = load_kimi_tokenizer_data(model_path)
-    tok = SimpleTokenizer(normal, special, "kimi-k2", bos_id=bos, eos_id=eos, eot_id=eos)
-    model_name = "Kimi-Linear-48B-A3B-Instruct-MXFP4"
-    ct = (model_path / "chat_template.jinja").read_text() if (model_path / "chat_template.jinja").exists() else None
-  else:
-    if args.devices != 1: raise ValueError("--devices is currently supported by the native Kimi MXFP4 loader only")
-    model, kv = Transformer.from_gguf(fetch(models.get(args.model, args.model)), args.max_context)
-    model_name = kv.get('general.name') or kv.get('general.basename') or args.model
-    tok = SimpleTokenizer.from_gguf_kv(kv)
-    ct = kv.get('tokenizer.chat_template')
+  model, kv = Transformer.from_gguf(fetch(models.get(args.model, args.model)), args.max_context)
+  model_name = kv.get('general.name') or kv.get('general.basename') or args.model
   file_sizes = [y.nbytes() for y in UOp.sink(*[x.uop for x in nn.state.get_parameters(model)]).toposort() if y.op is Ops.BUFFER]
   print(f"using model \"{model_name}\" with {sum(file_sizes):,} bytes and {sum(x.numel() for x in nn.state.get_parameters(model)):,} params, "
         f"max context {args.max_context} on {nn.state.get_parameters(model)[0].device}")
 
+  # get tokenizer
+  tok = SimpleTokenizer.from_gguf_kv(kv)
+
   # use the model's chat template if jinja2 is available (enables model-specific formatting)
-  template: jinja2.Template|FallbackTemplate|KimiK3Template = KimiK3Template() if is_k3 else FallbackTemplate(tok)
-  if ct is not None:
+  template: jinja2.Template|FallbackTemplate = FallbackTemplate(tok)
+  if (ct := kv.get('tokenizer.chat_template')) is not None:
     try:
       import jinja2
       env = jinja2.Environment()
@@ -227,10 +162,9 @@ def main():
       template = env.from_string(ct)
     except ImportError: print("warning: jinja2 is not installed, the model's chat template is disabled")
 
-  # Capture the default greedy serving shapes before accepting requests.
-  if args.warmup:
+  # warmup the JIT
+  if args.warmup or args.serve:
     with Context(DEBUG=max(DEBUG.value, 1)): model.warmup()
-  elif args.serve: model.warmup()
 
   # start server
   if args.serve: LLMServer(('', args.serve), model, model_name, tok, template).serve_forever()
@@ -255,26 +189,15 @@ def main():
   while 1:
     try: messages.append({"role":"user", "content":input('>>> ')})
     except EOFError: break
-    rendered = template.render(messages=messages, add_generation_prompt=True)
-    ids = tok.encode(rendered)
-    reply, reasoning_reply, dec = "", "", tok.stream_decoder()
-    xtml = rendered.rstrip().endswith("<|open|>think<|sep|>")
-    router = StreamRouter(reasoning=xtml or rendered.rstrip().endswith("<think>"), xtml=xtml)
+    ids = tok.encode(template.render(messages=messages, add_generation_prompt=True))
+    reply, dec = "", tok.stream_decoder()
     for next_id in model.generate(ids):
       if tok.is_end(next_id):
-        for field,text in router.route(dec(), final=True):
-          if field == "content": reply += text
-          elif field == "reasoning_content": reasoning_reply += text
-          sys.stdout.write(text)
-        sys.stdout.write("\n\n")
+        sys.stdout.write(dec() + "\n\n")
         break
-      for field,text in router.route(dec(next_id)):
-        if field == "content": reply += text
-        elif field == "reasoning_content": reasoning_reply += text
-        sys.stdout.write(text)
-        sys.stdout.flush()
-    assistant = {"role":"assistant", "content":reply}
-    if reasoning_reply: assistant["reasoning_content"] = reasoning_reply
-    messages.append(assistant)
+      reply += (piece := dec(next_id))
+      sys.stdout.write(piece)
+      sys.stdout.flush()
+    messages.append({"role":"assistant", "content":reply})
 
 if __name__ == "__main__": main()

tinygrad/llm/kernels/__init__.py

@@ -1,238 +0,0 @@
-import functools
-from typing import cast
-from tinygrad import Tensor, UOp, Device, Context, dtypes
-from tinygrad.device import Buffer, MultiBuffer
-from tinygrad.dtype import AddrSpace
-from tinygrad.uop.ops import AxisType, KernelInfo
-
-def amd_custom_kernels_supported(device:str|tuple[str, ...]|None) -> bool:
-  """The hand-written wave32 kernel is intentionally limited to RDNA3/gfx11."""
-  if device is None: return False
-  device = device[0] if isinstance(device, tuple) else device
-  with Context(ALLOW_DEVICE_USAGE=1):
-    return (target:=getattr(Device[device], "target", None)) is not None and target[0] == 11
-
-def amd_wave64_custom_kernels_supported(device:str|tuple[str, ...]|None) -> bool:
-  """CDNA4 wave64 kernels used by the MI350X K3 path."""
-  if device is None: return False
-  device = device[0] if isinstance(device, tuple) else device
-  if device.startswith("NULL:HIP:gfx950"): return True  # compile-only CDNA4 coverage without MI350X hardware
-  with Context(ALLOW_DEVICE_USAGE=1):
-    return (target:=getattr(Device[device], "target", None)) is not None and target[:2] == (9, 5)
-
-def amd_packed_mxfp4_supported(device:str|tuple[str, ...]|None) -> bool:
-  return amd_custom_kernels_supported(device) or amd_wave64_custom_kernels_supported(device)
-
-def amd_exact_bf16_custom_kernels_supported(device:str|tuple[str, ...]|None) -> bool:
-  """The LDS-reduced exact BF16 pair kernels are portable across RDNA3 and CDNA4."""
-  return amd_custom_kernels_supported(device) or amd_wave64_custom_kernels_supported(device)
-
-def amd_int32_item(x:Tensor, host:memoryview) -> int:
-  """Copy a realized replicated AMD scalar without constructing a new scheduler graph."""
-  if x.numel() != 1 or x.dtype != dtypes.int32 or host.nbytes != 4: raise ValueError("expected one int32 and a four-byte host view")
-  buf = x.uop.buffer
-  if isinstance(buf, MultiBuffer): buf = buf.bufs[0]
-  if not isinstance(buf, Buffer) or not buf.device.startswith("AMD"): raise ValueError("expected a realized AMD buffer")
-  buf.allocator._copyout(host, buf._buf)
-  return int.from_bytes(host, byteorder="little", signed=True)
-
-def mxfp4_expert_linear(sel:Tensor, x:Tensor, weight:Tensor, scale:Tensor, partial:bool=False) -> Tensor:
-  """Run a TP routed projection without materializing selected BF16 weights."""
-  from tinygrad.llm.kernels.amd import (_mxfp4_expert_linear_kernel, _mxfp4_expert_linear_wave64_kernel,
-                                        _mxfp4_expert_linear_wave64_prefill_kernel)
-  batch, tokens, topk = sel.shape
-  out_features = weight.shape[1]
-  weight_axis = weight.uop.axis
-  if isinstance(weight.device, tuple):
-    devices = weight.device
-    # Gate/up shard their output dimension. Down shards its reduction dimension;
-    # represent each GPU's partial as a size-one device axis, then all-reduce it.
-    axis = 3 if weight_axis == 1 else 4
-    shard_shape: tuple[int|UOp, ...]
-    if weight_axis == 1:
-      if out_features % len(devices): raise ValueError(f"expert output {out_features} is not divisible by {len(devices)} devices")
-      shard_shape = (batch, tokens, topk, out_features//len(devices))
-    elif weight_axis == 2:
-      shard_shape = (batch, tokens, topk, out_features, 1)
-    else: raise ValueError(f"unsupported expert TP axis {weight_axis}")
-    partial_dtype = dtypes.float32 if weight_axis == 2 else dtypes.bfloat16
-    parts = [Tensor.empty(*shard_shape, dtype=partial_dtype, device=device).uop for device in devices]
-    out = Tensor(parts[0].mstack(*parts[1:]).unshard(axis))
-  else:
-    out = Tensor.empty(batch, tokens, topk, out_features, dtype=dtypes.bfloat16, device=weight.device)
-  if amd_wave64_custom_kernels_supported(weight.device):
-    kernel = _mxfp4_expert_linear_wave64_prefill_kernel if tokens > 1 else _mxfp4_expert_linear_wave64_kernel
-  else: kernel = _mxfp4_expert_linear_kernel
-  out = Tensor.custom_kernel(out, sel.contiguous(), x.contiguous(), weight, scale, fxn=kernel)[0]
-  return out if weight_axis == 2 and partial else out.sum(4).cast(dtypes.bfloat16) if weight_axis == 2 else out
-
-def bf16_partial_linear(x:Tensor, weight:Tensor) -> Tensor:
-  """Return output-shaped FP32 TP partials with a final device axis, without all-reduce."""
-  from tinygrad.llm.kernels.amd import _bf16_partial_linear_kernel
-  if not isinstance(weight.device, tuple) or weight.uop.axis != 1: raise ValueError("partial linear expects input-sharded TP weight")
-  batch, tokens, _ = x.shape
-  devices, out_features = weight.device, weight.shape[0]
-  shard_shape = (batch, tokens, out_features, 1)
-  parts = [Tensor.empty(*shard_shape, dtype=dtypes.float32, device=device).uop for device in devices]
-  out = Tensor(parts[0].mstack(*parts[1:]).unshard(3))
-  return Tensor.custom_kernel(out, x.contiguous(), weight, fxn=_bf16_partial_linear_kernel)[0]
-
-def bf16_matvec(x:Tensor, weight:Tensor) -> Tensor:
-  from tinygrad.llm.kernels.amd import _bf16_matvec_kernel
-  batch, tokens, _ = x.shape
-  out_features = weight.shape[0]
-  if isinstance(weight.device, tuple):
-    devices = weight.device
-    if weight.uop.axis == 0:
-      shard_shape = (batch, tokens, out_features//len(devices))
-      parts = [Tensor.empty(*shard_shape, dtype=dtypes.bfloat16, device=device).uop for device in devices]
-      out = Tensor(parts[0].mstack(*parts[1:]).unshard(2))
-    elif weight.uop.axis is None: out = Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=weight.device)
-    else: raise ValueError("bf16_matvec expects output-sharded or replicated TP weight")
-  else: out = Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=weight.device)
-  return Tensor.custom_kernel(out, x.contiguous(), weight, fxn=_bf16_matvec_kernel)[0]
-
-def bf16_mfma_splitk(x:Tensor, weight:Tensor) -> Tensor:
-  """gfx950 decode matvec for replicated or output-sharded BF16 weights."""
-  from tinygrad.llm.kernels.amd import _bf16_mfma_splitk_kernel
-  batch, tokens, in_features = x.shape
-  out_features = weight.shape[0]
-  if batch != 1 or tokens != 1 or weight.shape[1] != in_features or in_features % 256:
-    raise ValueError(f"unsupported MFMA split-K shapes {x.shape} {weight.shape}")
-  if not amd_wave64_custom_kernels_supported(weight.device): raise ValueError("MFMA split-K requires gfx950")
-  if isinstance(weight.device, tuple):
-    devices = weight.device
-    if weight.uop.axis == 0:
-      if out_features % (16*len(devices)): raise ValueError("local MFMA output must be divisible by 16")
-      shape = (batch, tokens, out_features//len(devices))
-      parts = [Tensor.empty(*shape, dtype=dtypes.bfloat16, device=device).uop for device in devices]
-      out = Tensor(parts[0].mstack(*parts[1:]).unshard(2))
-    elif weight.uop.axis is None:
-      if out_features % 16: raise ValueError("MFMA output must be divisible by 16")
-      out = Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=devices)
-    else: raise ValueError("MFMA split-K requires replicated or output-sharded weight")
-  else:
-    if out_features % 16: raise ValueError("MFMA output must be divisible by 16")
-    out = Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=weight.device)
-  return Tensor.custom_kernel(out, x.contiguous(), weight, fxn=_bf16_mfma_splitk_kernel)[0]
-
-def mxfp8_quantize_dequantize(x:Tensor) -> Tensor:
-  """gfx11 software MXFP8 round trip without a multi-kernel reduction graph."""
-  from tinygrad.llm.kernels.amd import _mxfp8_qdq_kernel
-  out = Tensor.empty_like(x, dtype=dtypes.bfloat16)
-  return Tensor.custom_kernel(out, x.contiguous(), fxn=_mxfp8_qdq_kernel)[0]
-
-def kda_qkv_linear(x:Tensor, qw:Tensor, kw:Tensor, vw:Tensor) -> tuple[Tensor, Tensor, Tensor]:
-  """Fuse equal-sized output-sharded KDA Q/K/V decode projections."""
-  from tinygrad.llm.kernels.amd import _kda_qkv_kernel
-  batch, tokens, _ = x.shape
-  out_features = qw.shape[0]
-  if not (qw.shape == kw.shape == vw.shape): raise ValueError("fused KDA Q/K/V weights must have equal shapes")
-  if isinstance(qw.device, tuple):
-    devices = qw.device
-    if qw.uop.axis != 0 or out_features % len(devices): raise ValueError("fused KDA Q/K/V expects output-sharded weights")
-    shard_shape = (batch, tokens, out_features//len(devices))
-    def make_out() -> Tensor:
-      parts = [Tensor.empty(*shard_shape, dtype=dtypes.bfloat16, device=device).uop for device in devices]
-      return Tensor(parts[0].mstack(*parts[1:]).unshard(2))
-    outs: tuple[Tensor, Tensor, Tensor] = (make_out(), make_out(), make_out())
-  else:
-    outs = (Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=qw.device),
-            Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=qw.device),
-            Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=qw.device))
-  ret = Tensor.custom_kernel(*outs, x.contiguous(), qw, kw, vw, fxn=_kda_qkv_kernel)
-  return ret[0], ret[1], ret[2]
-
-def dual_bf16_matvec(x:Tensor, aw:Tensor, bw:Tensor, fast:bool=False) -> tuple[Tensor, Tensor]:
-  """Fuse two equal-shaped BF16 decode projections that consume the same input."""
-  from tinygrad.llm.kernels.amd import _dual_bf16_matvec_fast_kernel, _dual_bf16_matvec_kernel
-  batch, tokens, _ = x.shape
-  out_features = aw.shape[0]
-  if aw.shape != bw.shape: raise ValueError("fused BF16 weights must have equal shapes")
-  if isinstance(aw.device, tuple) and aw.uop.axis == 0:
-    devices = aw.device
-    if out_features % len(devices): raise ValueError("fused BF16 output is not divisible by the device count")
-    shard_shape = (batch, tokens, out_features//len(devices))
-    def make_out() -> Tensor:
-      parts = [Tensor.empty(*shard_shape, dtype=dtypes.bfloat16, device=device).uop for device in devices]
-      return Tensor(parts[0].mstack(*parts[1:]).unshard(2))
-    outs = (make_out(), make_out())
-  else:
-    outs = (Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=aw.device),
-            Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=aw.device))
-  ret = Tensor.custom_kernel(*outs, x.contiguous(), aw, bw, fxn=_dual_bf16_matvec_fast_kernel if fast else _dual_bf16_matvec_kernel)
-  return ret[0], ret[1]
-
-def dual_input_bf16_matvec(ax:Tensor, bx:Tensor, aw:Tensor, bw:Tensor) -> tuple[Tensor, Tensor]:
-  """Fuse equal-shaped BF16 projections with separate inputs and identical TP layouts."""
-  from tinygrad.llm.kernels.amd import _dual_input_bf16_matvec_kernel
-  if ax.shape != bx.shape or aw.shape != bw.shape: raise ValueError("fused BF16 inputs and weights must have equal shapes")
-  batch, tokens, _ = ax.shape
-  out_features = aw.shape[0]
-  if isinstance(aw.device, tuple) and aw.uop.axis == 0:
-    devices = aw.device
-    if out_features % len(devices): raise ValueError("fused BF16 output is not divisible by the device count")
-    shard_shape = (batch, tokens, out_features//len(devices))
-    def make_out() -> Tensor:
-      parts = [Tensor.empty(*shard_shape, dtype=dtypes.bfloat16, device=device).uop for device in devices]
-      return Tensor(parts[0].mstack(*parts[1:]).unshard(2))
-    outs = (make_out(), make_out())
-  else:
-    outs = (Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=aw.device),
-            Tensor.empty(batch, tokens, out_features, dtype=dtypes.bfloat16, device=aw.device))
-  ret = Tensor.custom_kernel(*outs, ax.contiguous(), bx.contiguous(), aw, bw, fxn=_dual_input_bf16_matvec_kernel)
-  return ret[0], ret[1]
-
-def kda_fgb_linear(x:Tensor, gw:Tensor, fw:Tensor, bw:Tensor) -> tuple[Tensor, Tensor, Tensor]:
-  """Fuse replicated KDA g/f low-rank projections with its output-sharded beta projection."""
-  from tinygrad.llm.kernels.amd import _kda_fgb_kernel
-  if gw.shape != fw.shape or not isinstance(gw.device, tuple) or gw.uop.axis is not None or \
-     bw.device != gw.device or bw.uop.axis != 0: raise ValueError("unsupported KDA f/g/beta TP layout")
-  batch, tokens, _ = x.shape
-  devices, rank, beta_features = gw.device, gw.shape[0], bw.shape[0]
-  gout = Tensor.empty(batch, tokens, rank, dtype=dtypes.bfloat16, device=devices)
-  fout = Tensor.empty(batch, tokens, rank, dtype=dtypes.bfloat16, device=devices)
-  beta_shape = (batch, tokens, beta_features//len(devices))
-  parts = [Tensor.empty(*beta_shape, dtype=dtypes.bfloat16, device=device).uop for device in devices]
-  bout = Tensor(parts[0].mstack(*parts[1:]).unshard(2))
-  ret = Tensor.custom_kernel(gout, fout, bout, x.contiguous(), gw, fw, bw, fxn=_kda_fgb_kernel)
-  return ret[0], ret[1], ret[2]
-
-@functools.cache
-def _gated_delta_prefill_kernel(core:UOp, next_state:UOp, q:UOp, k:UOp, v:UOp, beta:UOp, alpha:UOp, state:UOp, kq:UOp) -> UOp:
-  batch, heads, tokens, value_dim = cast(tuple[int, int, int, int], core.shape)
-  key_dim, alpha_dim = cast(int, q.shape[-1]), cast(int, alpha.shape[-1]) if len(alpha.shape) == 4 else 1
-  core, v = (x.reshape(batch*heads, tokens, value_dim) for x in (core, v))
-  q, k = (x.reshape(batch*heads, tokens, key_dim) for x in (q, k))
-  beta, kq = (x.reshape(batch*heads, tokens) for x in (beta, kq))
-  alpha = alpha.reshape(batch*heads, tokens, alpha_dim)
-  state, next_state = (x.reshape(batch*heads, value_dim, key_dim) for x in (state, next_state))
-  bh, row, cols = UOp.range(batch*heads, 0, AxisType.GLOBAL), UOp.range(value_dim, 2), tuple(range(key_dim))
-  current = UOp.placeholder((key_dim,), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  current = current.after(UOp.group(*(current[col].store(state[bh, row, col].float()) for col in cols)))
-  token = UOp.range(tokens, 1, AxisType.REDUCE)
-  previous = tuple(current.after(token)[col].load() for col in cols)
-  keys, queries = (tuple(x[bh, token, col].load() for col in cols) for x in (k, q))
-  av = tuple(alpha[bh, token, col if alpha_dim > 1 else 0].load() for col in cols)
-  bv = beta[bh, token].load()
-  state_k = sum((x*a*y for x,a,y in zip(previous, av, keys)), UOp.const(0, dtypes.float32))
-  state_q = sum((x*a*y for x,a,y in zip(previous, av, queries)), UOp.const(0, dtypes.float32))
-  delta = (v[bh, token, row].load() - state_k) * bv
-  step = UOp.group(core[bh, token, row].store(state_q + delta*kq[bh, token]),
-                   *(current[col].store(x*a + delta*y) for col,x,a,y in zip(cols, previous, av, keys))).end(token)
-  stores = (next_state[bh, row, col].store(current.after(step)[col].load().cast(next_state.dtype)) for col in cols)
-  return UOp.group(*stores).end(row, bh).sink(arg=KernelInfo(name="gated_delta_prefill", opts_to_apply=()))
-
-def gated_delta_prefill(q:Tensor, k:Tensor, v:Tensor, beta:Tensor, alpha:Tensor, state:Tensor) -> tuple[Tensor, Tensor]:
-  batch, heads, tokens, key_dim = q.shape
-  value_dim = v.shape[-1]
-  assert q.shape == k.shape and v.shape[:3] == q.shape[:3] and beta.shape == (batch, heads, tokens)
-  assert alpha.shape in ((batch, heads, tokens), (batch, heads, tokens, key_dim))
-  assert state.shape == (batch, heads, value_dim, key_dim)
-  kernel = _gated_delta_prefill_kernel
-  if amd_exact_bf16_custom_kernels_supported(q.device) and key_dim % 32 == 0 and value_dim % 4 == 0:
-    from tinygrad.llm.kernels.amd import _gated_delta_prefill_kernel as kernel
-  core, next_state, kq = Tensor.empty_like(v), Tensor.empty_like(state), (q*k).sum(-1).contiguous()
-  result = Tensor.custom_kernel(core, next_state, q.contiguous(), k.contiguous(), v.contiguous(), beta.contiguous(), alpha.contiguous(), state, kq,
-                                fxn=kernel)
-  return result[0], result[1]

tinygrad/llm/kernels/amd.py

@@ -1,353 +0,0 @@
-from __future__ import annotations
-import functools
-import pathlib
-from typing import cast
-from tinygrad import UOp
-from tinygrad.uop.ops import AxisType, KernelInfo, Ops
-from tinygrad.dtype import AddrSpace, dtypes
-
-@functools.cache
-def _bf16_mfma_splitk_kernel(out:UOp, x:UOp, weight:UOp) -> UOp:
-  """CDNA4 BF16 matvec with eight waves splitting K per 16 output channels."""
-  from tinygrad.renderer import Estimates
-  from tinygrad.runtime.support.compiler_amd import HIPCCCompiler
-  out_features, in_features = cast(tuple[int, int], weight.shape)
-  assert out.numel() == out_features and x.numel() == in_features and out_features % 16 == 0 and in_features % 256 == 0
-  threads, workgroups = UOp.special(512, "lidx0"), UOp.special(out_features//16, "gidx0")
-  sink = UOp.sink(out.base, x.base, weight.base, threads, workgroups,
-                  arg=KernelInfo(name=f"bf16_mfma_splitk_{out_features}_{in_features}",
-                                 estimates=Estimates(ops=2*out_features*in_features,
-                                                     mem=(out_features*in_features+in_features+out_features)*2)))
-  root = pathlib.Path(__file__).parents[3]/"extra"/"thunder"/"amd"
-  src = (root/"matvec_bf16_splitk.cpp").read_text()
-  lib = HIPCCCompiler("gfx950", [f"-I{(root/'include').as_posix()}", "-std=c++20", "-DKITTENS_CDNA4", "-ffast-math",
-                                   "-DHIP_ENABLE_WARP_SYNC_BUILTINS", f"-DMATVEC_N={out_features}",
-                                   f"-DMATVEC_K={in_features}"]).compile_cached(src)
-  return UOp(Ops.PROGRAM, src=(sink, UOp(Ops.LINEAR, src=(*sink.src, sink)), UOp(Ops.SOURCE, arg=src), UOp(Ops.BINARY, arg=lib)))
-
-def warp_reduce(val:UOp, full_wave:bool=False, maximum:bool=False) -> UOp:
-  for offset in ((16, 8, 4, 2, 1) if full_wave else (8, 4, 2, 1)):
-    if val.op is Ops.INDEX and val.addrspace == AddrSpace.REG: val = val.load()
-    other = UOp(Ops.CUSTOM, dtypes.float, (val,), arg=
-      f"__builtin_bit_cast(float, __builtin_amdgcn_ds_swizzle(__builtin_bit_cast(int, {{0}}), {0x1f | offset<<10}))")
-    val = val.maximum(other) if maximum else val + other
-  return val
-
-@functools.cache
-def _mxfp8_qdq_kernel(out:UOp, x:UOp) -> UOp:
-  """Software OCP E4M3/E8M0 round trip, one wave per 32-value MX block."""
-  groups = cast(int, x.shape[-1])//32
-  outer = x.numel()//cast(int, x.shape[-1])
-  block, lane = UOp.range(outer*groups, 0), UOp.range(32, 1, axis_type=AxisType.LOCAL)
-  value = x.reshape(outer, groups, 32)[block//groups, block%groups, lane].float()
-  amax = warp_reduce(value.abs(), full_wave=True, maximum=True)
-  exponent = (amax.maximum(1e-38)/448.0).log2().round().maximum(-127.0).minimum(127.0)
-  block_scale = amax.eq(0).where(1.0, exponent.exp2())
-  normalized = value/block_scale
-  magnitude = normalized.abs().minimum(448.0)
-  elem_exp = magnitude.maximum(2**-9).log2().floor().maximum(-6.0).minimum(8.0)
-  quantum = (elem_exp-3.0).exp2()
-  quantized = (magnitude/quantum).round()*quantum
-  quantized = (normalized < 0).where(-quantized, quantized).maximum(-448.0).minimum(448.0)
-  store = out.reshape(outer, groups, 32)[block//groups, block%groups, lane].store((quantized*block_scale).cast(out.dtype))
-  return store.end(lane, block).sink(arg=KernelInfo(name="mxfp8_qdq", opts_to_apply=()))
-
-def _mxfp4_value(code:UOp) -> UOp:
-  """Decode one OCP E2M1 nibble without a lookup-table memory access."""
-  magnitude = code & 7
-  value = magnitude.eq(7).where(6.0, magnitude.eq(6).where(4.0, magnitude.eq(5).where(3.0, magnitude.float()*0.5)))
-  return (code & 8).ne(0).where(-value, value)
-
-def _e8m0_value(scale:UOp) -> UOp:
-  """Decode an E8M0 byte with IEEE exponent bits instead of a transcendental exp2."""
-  bits = scale.cast(dtypes.uint32) << 23
-  # E8M0 byte zero denotes 2**-127, halfway through IEEE's subnormal exponent bin.
-  return scale.eq(0).where(UOp.const(0x00400000, dtypes.uint32).bitcast(dtypes.float32), bits.bitcast(dtypes.float32))
-
-@functools.cache
-def _kda_qkv_kernel(qout:UOp, kout:UOp, vout:UOp, x:UOp, qw:UOp, kw:UOp, vw:UOp) -> UOp:
-  """Fused BF16 decode projection for equal-sized KDA Q/K/V tensors."""
-  batch, tokens, out_features = cast(tuple[int, int, int], qout.shape)
-  in_features, output_tile = cast(int, x.shape[-1]), 1
-  assert qout.shape == kout.shape == vout.shape and out_features % output_tile == 0 and in_features % 32 == 0
-  row, lane = UOp.range(batch*tokens*(out_features//output_tile), 0), UOp.range(32, 1, axis_type=AxisType.LOCAL)
-  token, output_block = row // (out_features//output_tile), row % (out_features//output_tile)
-  outputs = tuple(output_block*output_tile+i for i in range(output_tile))
-  acc = UOp.placeholder((3, output_tile), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  acc = acc.after(acc.store(acc.const_like(0.0)))
-  group = UOp.range(in_features//32, 2, AxisType.REDUCE)
-  activation = x.reshape(batch*tokens, in_features)[token, group*32+lane].float()
-  updates = [acc.after(group)[p, i].load()+activation*w[output, group*32+lane].float()
-             for p,w in enumerate((qw, kw, vw)) for i,output in enumerate(outputs)]
-  update = acc.store(UOp.stack(*updates).reshape(3, output_tile)).end(group)
-  outs = (qout, kout, vout)
-  stores = (outs[p].reshape(batch*tokens, out_features)[token, output.valid(lane.eq(0))].store(
-    warp_reduce(acc.after(update)[p, i], full_wave=True).cast(outs[p].dtype))
-    for p in range(3) for i,output in enumerate(outputs))
-  return UOp.group(*stores).end(lane, row).sink(arg=KernelInfo(name="kda_qkv", opts_to_apply=()))
-
-@functools.cache
-def _dual_bf16_matvec_kernel(aout:UOp, bout:UOp, x:UOp, aw:UOp, bw:UOp) -> UOp:
-  """Exact pair of BF16 decode projections with one activation read."""
-  batch, tokens, out_features = cast(tuple[int, int, int], aout.shape)
-  in_features = cast(int, x.shape[-1])
-  assert aout.shape == bout.shape and out_features == aw.shape[0] == bw.shape[0] and in_features % 16 == 0
-  row, lane = UOp.range(batch*tokens*out_features, 0), UOp.range(16, 1, axis_type=AxisType.LOCAL)
-  token, output = row//out_features, row%out_features
-  acc = UOp.placeholder((2,), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  acc = acc.after(acc.store(acc.const_like(0.0)))
-  chunk, group = in_features//16, UOp.range(in_features//16, 2, AxisType.REDUCE)
-  input_idx = lane*chunk+group
-  activation = x.reshape(batch*tokens, in_features)[token, input_idx].float()
-  update = acc.store(UOp.stack(*(acc.after(group)[i].load()+(activation*w[output, input_idx].float()).cast(dtypes.bfloat16).float()
-                                for i,w in enumerate((aw, bw))))).end(group)
-  local = UOp.placeholder((2, 16), dtypes.float32, slot=1, addrspace=AddrSpace.LOCAL)
-  barrier = UOp.group(*(local[i, lane].store(acc.after(update)[i]) for i in range(2))).barrier()
-  stores = (out.reshape(batch*tokens, out_features)[token, output.valid(lane.eq(0))].store(
-    sum((local.after(barrier)[i, j] for j in range(16)), UOp.const(0, dtypes.float32)).cast(out.dtype))
-            for i,out in enumerate((aout, bout)))
-  return UOp.group(*stores).end(lane, row).sink(arg=KernelInfo(name="dual_bf16_matvec", opts_to_apply=()))
-
-@functools.cache
-def _dual_bf16_matvec_fast_kernel(aout:UOp, bout:UOp, x:UOp, aw:UOp, bw:UOp) -> UOp:
-  """Coalesced pair used where one changed reduction boundary does not feed recurrent state."""
-  batch, tokens, out_features = cast(tuple[int, int, int], aout.shape)
-  in_features = cast(int, x.shape[-1])
-  assert aout.shape == bout.shape and out_features == aw.shape[0] == bw.shape[0] and in_features % 32 == 0
-  row, lane = UOp.range(batch*tokens*out_features, 0), UOp.range(32, 1, axis_type=AxisType.LOCAL)
-  token, output = row//out_features, row%out_features
-  acc = UOp.placeholder((2,), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  acc = acc.after(acc.store(acc.const_like(0.0)))
-  group = UOp.range(in_features//32, 2, AxisType.REDUCE)
-  input_idx = group*32+lane
-  activation = x.reshape(batch*tokens, in_features)[token, input_idx].float()
-  update = acc.store(UOp.stack(*(acc.after(group)[i].load()+(activation*w[output, input_idx].float()).cast(dtypes.bfloat16).float()
-                                for i,w in enumerate((aw, bw))))).end(group)
-  stores = (out.reshape(batch*tokens, out_features)[token, output.valid(lane.eq(0))].store(
-    warp_reduce(acc.after(update)[i], full_wave=True).cast(out.dtype)) for i,out in enumerate((aout, bout)))
-  return UOp.group(*stores).end(lane, row).sink(arg=KernelInfo(name="dual_bf16_matvec_fast", opts_to_apply=()))
-
-@functools.cache
-def _dual_input_bf16_matvec_kernel(aout:UOp, bout:UOp, ax:UOp, bx:UOp, aw:UOp, bw:UOp) -> UOp:
-  """Exact pair of equal-shaped BF16 projections with distinct inputs."""
-  batch, tokens, out_features = cast(tuple[int, int, int], aout.shape)
-  in_features = cast(int, ax.shape[-1])
-  assert aout.shape == bout.shape and ax.shape == bx.shape and out_features == aw.shape[0] == bw.shape[0] and in_features % 16 == 0
-  row, lane = UOp.range(batch*tokens*out_features, 0), UOp.range(16, 1, axis_type=AxisType.LOCAL)
-  token, output = row//out_features, row%out_features
-  acc = UOp.placeholder((2,), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  acc = acc.after(acc.store(acc.const_like(0.0)))
-  chunk, group = in_features//16, UOp.range(in_features//16, 2, AxisType.REDUCE)
-  input_idx = lane*chunk+group
-  update = acc.store(UOp.stack(*(acc.after(group)[i].load()+(inp.reshape(batch*tokens, in_features)[token, input_idx].float()*
-    weight[output, input_idx].float()).cast(dtypes.bfloat16).float() for i,(inp,weight) in enumerate(((ax,aw), (bx,bw)))))).end(group)
-  local = UOp.placeholder((2, 16), dtypes.float32, slot=1, addrspace=AddrSpace.LOCAL)
-  barrier = UOp.group(*(local[i, lane].store(acc.after(update)[i]) for i in range(2))).barrier()
-  stores = (out.reshape(batch*tokens, out_features)[token, output.valid(lane.eq(0))].store(
-    sum((local.after(barrier)[i, j] for j in range(16)), UOp.const(0, dtypes.float32)).cast(out.dtype))
-            for i,out in enumerate((aout, bout)))
-  return UOp.group(*stores).end(lane, row).sink(arg=KernelInfo(name="dual_input_bf16_matvec", opts_to_apply=()))
-
-@functools.cache
-def _kda_fgb_kernel(gout:UOp, fout:UOp, bout:UOp, x:UOp, gw:UOp, fw:UOp, bw:UOp) -> UOp:
-  """Mixed-output wave32 KDA g/f/beta projection."""
-  batch, tokens, rank = cast(tuple[int, int, int], gout.shape)
-  beta_features, in_features = cast(int, bout.shape[-1]), cast(int, x.shape[-1])
-  assert gout.shape == fout.shape and rank == gw.shape[0] == fw.shape[0] and beta_features == bw.shape[0] and in_features % 32 == 0
-  rows = batch*tokens*(2*rank+beta_features)
-  row, lane = UOp.range(rows, 0), UOp.range(32, 1, axis_type=AxisType.LOCAL)
-  token, projection_row = row//(2*rank+beta_features), row%(2*rank+beta_features)
-  is_g, is_f = projection_row < rank, (projection_row >= rank) & (projection_row < 2*rank)
-  g_row = projection_row.valid(is_g)
-  f_row = (projection_row-rank).valid(is_f)
-  b_row = (projection_row-2*rank).valid(~is_g & ~is_f)
-  acc = UOp.placeholder((), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  acc = acc.after(acc.store(0.0))
-  group = UOp.range(in_features//32, 2, AxisType.REDUCE)
-  input_idx = group*32+lane
-  weight = is_g.where(gw[g_row, input_idx], is_f.where(fw[f_row, input_idx], bw[b_row, input_idx])).float()
-  product = (x.reshape(batch*tokens, in_features)[token, input_idx].float()*weight).cast(dtypes.bfloat16).float()
-  update = acc.store(acc.after(group)+product).end(group)
-  total = warp_reduce(acc.after(update)[0], full_wave=True).cast(dtypes.bfloat16)
-  stores = (gout.reshape(batch*tokens, rank)[token, g_row.valid(lane.eq(0))].store(total),
-            fout.reshape(batch*tokens, rank)[token, f_row.valid(lane.eq(0))].store(total),
-            bout.reshape(batch*tokens, beta_features)[token, b_row.valid(lane.eq(0))].store(total))
-  return UOp.group(*stores).end(lane, row).sink(arg=KernelInfo(name="kda_fgb", opts_to_apply=()))
-
-def _mxfp4_expert_linear_impl(out:UOp, sel:UOp, x:UOp, weight:UOp, scale:UOp) -> UOp:
-  """Wave32 decode GEMM which consumes selected experts directly from packed MXFP4 storage."""
-  batch, tokens, topk, out_features = cast(tuple[int, int, int, int], out.shape[:4])
-  partials = cast(int, out.shape[4]) if len(out.shape) == 5 else 1
-  output_tile = 1
-  assert out_features % output_tile == 0
-  in_features = cast(int, weight.shape[-1])*2
-  assert in_features % 32 == 0 and x.shape[-1] == in_features and sel.shape == (batch, tokens, topk)
-  xchoices = cast(int, x.shape[-2])
-  assert xchoices in (1, topk)
-  total_rows = batch*tokens*topk*(out_features//output_tile)*partials
-  row, lane = UOp.range(total_rows, 0), UOp.range(32, 1, axis_type=AxisType.LOCAL)
-  partial, output_block, route = row % partials, (row//partials) % (out_features//output_tile), \
-    row // ((out_features//output_tile)*partials)
-  outputs = tuple(output_block*output_tile+i for i in range(output_tile))
-  token, choice = route // topk, route % topk
-  expert = sel.reshape(batch*tokens, topk)[token, choice]
-  xv = x.reshape(batch*tokens, xchoices, in_features)
-  acc = UOp.placeholder((output_tile,), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  acc = acc.after(acc.store(acc.const_like(0.0)))
-  group = UOp.range(in_features//32, 2, AxisType.REDUCE)
-  activation = xv[token, 0 if xchoices == 1 else choice, group*32+lane].float()
-  updates = []
-  for i,output in enumerate(outputs):
-    packed = weight[expert, output, group*16+lane//2]
-    code = (packed >> ((lane&1)*4).cast(dtypes.uint8)) & 15
-    w = _mxfp4_value(code) * _e8m0_value(scale[expert, output, group])
-    updates.append(acc.after(group)[i].load()+activation*w)
-  update = acc.store(UOp.stack(*updates)).end(group)
-  out = out.reshape(batch*tokens, topk, out_features, partials)
-  stores = (out[token, choice, output, partial.valid(lane.eq(0))].store(warp_reduce(acc.after(update)[i], full_wave=True).cast(out.dtype))
-            for i,output in enumerate(outputs))
-  return UOp.group(*stores).end(lane, row).sink(arg=KernelInfo(name="mxfp4_expert_linear", opts_to_apply=()))
-
-@functools.cache
-def _mxfp4_expert_linear_kernel(out:UOp, sel:UOp, x:UOp, weight:UOp, scale:UOp) -> UOp:
-  return _mxfp4_expert_linear_impl(out, sel, x, weight, scale)
-
-@functools.cache
-def _mxfp4_expert_linear_wave64_kernel(out:UOp, sel:UOp, x:UOp, weight:UOp, scale:UOp) -> UOp:
-  """Wave64 decode GEMM for CDNA, with a workgroup-wide reduction independent of local-id decomposition."""
-  batch, tokens, topk, out_features = cast(tuple[int, int, int, int], out.shape[:4])
-  partials = cast(int, out.shape[4]) if len(out.shape) == 5 else 1
-  in_features = cast(int, weight.shape[-1])*2
-  assert in_features % 64 == 0 and x.shape[-1] == in_features and sel.shape == (batch, tokens, topk)
-  xchoices = cast(int, x.shape[-2])
-  assert xchoices in (1, topk)
-  total_rows = batch*tokens*topk*out_features*partials
-  row, lane = UOp.range(total_rows, 0), UOp.range(64, 1, axis_type=AxisType.LOCAL)
-  partial, output, route = row%partials, (row//partials)%out_features, row//(out_features*partials)
-  token, choice = route//topk, route%topk
-  expert = sel.reshape(batch*tokens, topk)[token, choice]
-  xv = x.reshape(batch*tokens, xchoices, in_features)
-  acc = UOp.placeholder((), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  acc = acc.after(acc.store(0.0))
-  group = UOp.range(in_features//64, 2, AxisType.REDUCE)
-  activation = xv[token, 0 if xchoices == 1 else choice, group*64+lane].float()
-  packed = weight[expert, output, group*32+lane//2]
-  code = (packed >> ((lane&1)*4).cast(dtypes.uint8)) & 15
-  weight_value = _mxfp4_value(code) * _e8m0_value(scale[expert, output, group*2+lane//32])
-  update = acc.store(acc.after(group)+activation*weight_value).end(group)
-  local = UOp.placeholder((64,), dtypes.float32, slot=1, addrspace=AddrSpace.LOCAL)
-  barrier = local[lane].store(acc.after(update)[0]).barrier()
-  total = sum((local.after(barrier)[i] for i in range(64)), UOp.const(0, dtypes.float32))
-  out = out.reshape(batch*tokens, topk, out_features, partials)
-  store = out[token, choice, output, partial.valid(lane.eq(0))].store(total.cast(out.dtype))
-  return store.end(lane, row).sink(arg=KernelInfo(name="mxfp4_expert_linear_wave64", opts_to_apply=()))
-
-@functools.cache
-def _mxfp4_expert_linear_wave64_prefill_kernel(out:UOp, sel:UOp, x:UOp, weight:UOp, scale:UOp) -> UOp:
-  """Tiled CDNA4 prefill GEMM. Four adjacent outputs share activation loads and each wave half reduces in parallel."""
-  batch, tokens, topk, out_features = cast(tuple[int, int, int, int], out.shape[:4])
-  partials = cast(int, out.shape[4]) if len(out.shape) == 5 else 1
-  in_features, output_tile = cast(int, weight.shape[-1])*2, 4
-  assert tokens > 1 and in_features % 64 == 0 and out_features % output_tile == 0
-  assert x.shape[-1] == in_features and sel.shape == (batch, tokens, topk)
-  xchoices = cast(int, x.shape[-2])
-  assert xchoices in (1, topk)
-  total_rows = batch*tokens*topk*(out_features//output_tile)*partials
-  row, lane = UOp.range(total_rows, 0), UOp.range(64, 1, axis_type=AxisType.LOCAL)
-  partial, output_block, route = row%partials, (row//partials)%(out_features//output_tile), row//((out_features//output_tile)*partials)
-  outputs = tuple(output_block*output_tile+i for i in range(output_tile))
-  token, choice = route//topk, route%topk
-  expert = sel.reshape(batch*tokens, topk)[token, choice]
-  xv = x.reshape(batch*tokens, xchoices, in_features)
-  acc = UOp.placeholder((output_tile,), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  acc = acc.after(acc.store(acc.const_like(0.0)))
-  group = UOp.range(in_features//64, 2, AxisType.REDUCE)
-  activation = xv[token, 0 if xchoices == 1 else choice, group*64+lane].float()
-  updates = []
-  for i,output in enumerate(outputs):
-    packed = weight[expert, output, group*32+lane//2]
-    code = (packed >> ((lane&1)*4).cast(dtypes.uint8)) & 15
-    weight_value = _mxfp4_value(code) * _e8m0_value(scale[expert, output, group*2+lane//32])
-    updates.append(acc.after(group)[i].load()+activation*weight_value)
-  update = acc.store(UOp.stack(*updates)).end(group)
-  half_totals = tuple(warp_reduce(acc.after(update)[i], full_wave=True) for i in range(output_tile))
-  local = UOp.placeholder((output_tile, 2), dtypes.float32, slot=1, addrspace=AddrSpace.LOCAL)
-  half = (lane//32).valid((lane&31).eq(0))
-  barrier = UOp.group(*(local[i, half].store(total) for i,total in enumerate(half_totals))).barrier()
-  out = out.reshape(batch*tokens, topk, out_features, partials)
-  stores = (out[token, choice, output, partial.valid(lane.eq(0))].store(
-    (local.after(barrier)[i, 0]+local.after(barrier)[i, 1]).cast(out.dtype)) for i,output in enumerate(outputs))
-  return UOp.group(*stores).end(lane, row).sink(arg=KernelInfo(name="mxfp4_expert_linear_wave64_prefill", opts_to_apply=()))
-
-@functools.cache
-def _bf16_partial_linear_kernel(out:UOp, x:UOp, weight:UOp) -> UOp:
-  """Per-device BF16 down projection; its dummy final axis is reduced after combining TP partials."""
-  batch, tokens, out_features, partials = cast(tuple[int, int, int, int], out.shape)
-  in_features, output_tile = cast(int, x.shape[-1]), 1
-  assert out_features % output_tile == 0 and in_features % 32 == 0
-  row, lane = UOp.range(batch*tokens*(out_features//output_tile)*partials, 0), UOp.range(32, 1, axis_type=AxisType.LOCAL)
-  partial, output_block, token = row%partials, (row//partials)%(out_features//output_tile), row//(partials*(out_features//output_tile))
-  outputs = tuple(output_block*output_tile+i for i in range(output_tile))
-  acc = UOp.placeholder((output_tile,), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  acc = acc.after(acc.store(acc.const_like(0.0)))
-  group = UOp.range(in_features//32, 2, AxisType.REDUCE)
-  input_idx = group*32+lane
-  activation = x.reshape(batch*tokens, in_features)[token, input_idx].float()
-  update = acc.store(UOp.stack(*(acc.after(group)[i].load()+(activation*weight[output, input_idx].float()).cast(dtypes.bfloat16).float()
-    for i,output in enumerate(outputs)))).end(group)
-  stores = (out.reshape(batch*tokens, out_features, partials)[token, output, partial.valid(lane.eq(0))].store(
-    warp_reduce(acc.after(update)[i], full_wave=True)) for i,output in enumerate(outputs))
-  return UOp.group(*stores).end(lane, row).sink(arg=KernelInfo(name="bf16_partial_linear", opts_to_apply=()))
-
-@functools.cache
-def _bf16_matvec_kernel(out:UOp, x:UOp, weight:UOp) -> UOp:
-  batch, tokens, out_features = cast(tuple[int, int, int], out.shape)
-  in_features = cast(int, x.shape[-1])
-  assert in_features % 32 == 0
-  row, lane = UOp.range(batch*tokens*out_features, 0), UOp.range(32, 1, axis_type=AxisType.LOCAL)
-  token, output = row//out_features, row%out_features
-  acc = UOp.placeholder((), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  acc = acc.after(acc.store(0.0))
-  group = UOp.range(in_features//32, 2, AxisType.REDUCE)
-  input_idx = group*32+lane
-  product = (x.reshape(batch*tokens, in_features)[token, input_idx].float()*weight[output, input_idx].float()).cast(dtypes.bfloat16).float()
-  update = acc.store(acc.after(group)+product).end(group)
-  total = warp_reduce(acc.after(update)[0], full_wave=True)
-  return out.reshape(batch*tokens, out_features)[token, output.valid(lane.eq(0))].store(total.cast(out.dtype)).end(lane, row).sink(
-    arg=KernelInfo(name="bf16_matvec", opts_to_apply=()))
-
-@functools.cache
-def _gated_delta_prefill_kernel(core:UOp, next_state:UOp, q:UOp, k:UOp, v:UOp, beta:UOp, alpha:UOp, state:UOp, kq:UOp) -> UOp:
-  batch, heads, tokens, value_dim, row_tile = *core.shape, 4
-  key_dim, alpha_dim = q.shape[-1], alpha.shape[-1] if len(alpha.shape) == 4 else 1
-  assert all(isinstance(x, int) for x in (batch, heads, tokens, value_dim, key_dim)) and key_dim % 32 == 0 and value_dim % row_tile == 0
-  batch, heads, tokens, value_dim, key_dim = cast(tuple[int, int, int, int, int], (batch, heads, tokens, value_dim, key_dim))
-  core, v = (x.reshape(batch*heads, tokens, value_dim) for x in (core, v))
-  q, k = (x.reshape(batch*heads, tokens, key_dim) for x in (q, k))
-  beta, kq = (x.reshape(batch*heads, tokens) for x in (beta, kq))
-  alpha = alpha.reshape(batch*heads, tokens, alpha_dim)
-  state, next_state = (x.reshape(batch*heads, value_dim, key_dim) for x in (state, next_state))
-  bh_row, lane = UOp.range(batch*heads*value_dim//row_tile, 0), UOp.range(32, 1, axis_type=AxisType.LOCAL)
-  bh, row_base = bh_row // (value_dim//row_tile), (bh_row % (value_dim//row_tile))*row_tile
-  rows = tuple(row_base+i for i in range(row_tile))
-  cols = tuple(lane + i*32 for i in range(key_dim//32))
-  current = UOp.placeholder((row_tile*key_dim//32,), dtypes.float32, slot=0, addrspace=AddrSpace.REG)
-  current = current.after(current.store(UOp.stack(*(state[bh, row, col].float() for row in rows for col in cols))))
-  token = UOp.range(tokens, 2, AxisType.REDUCE)
-  keys = tuple(k[bh, token, col].load() for col in cols)
-  queries = tuple(q[bh, token, col].load() for col in cols)
-  updates:list[UOp] = []
-  stores:list[UOp] = []
-  for row_idx,row in enumerate(rows):
-    previous = tuple(current.after(token)[row_idx*key_dim//32+i].load() for i in range(key_dim//32))
-    av = tuple(alpha[bh, token, col if alpha_dim > 1 else 0].load() for col in cols)
-    bv = beta[bh, token].load()
-    state_k = warp_reduce(sum((x*a*y for x,a,y in zip(previous, av, keys)), UOp.const(0, dtypes.float32)), full_wave=True)
-    state_q = warp_reduce(sum((x*a*y for x,a,y in zip(previous, av, queries)), UOp.const(0, dtypes.float32)), full_wave=True)
-    delta = (v[bh, token, row].load() - state_k) * bv
-    updates += [x*a + delta*y for x,a,y in zip(previous, av, keys)]
-    stores.append(core[bh, token, row.valid(lane.eq(0))].store(state_q + delta*kq[bh, token]))
-  step = UOp.group(*stores, current.store(UOp.stack(*updates))).end(token)
-  state_stores = (next_state[bh, row, col].store(current.after(step)[row_idx*key_dim//32+i].load().cast(next_state.dtype))
-                  for row_idx,row in enumerate(rows) for i,col in enumerate(cols))
-  return UOp.group(*state_stores).end(lane, bh_row).sink(arg=KernelInfo(name="gated_delta_prefill", opts_to_apply=()))

tinygrad/llm/kimi.py

@@ -1,181 +0,0 @@
-from __future__ import annotations
-import base64, gc, json, pathlib, shutil
-from tinygrad import Tensor, Device, dtypes, nn
-from tinygrad.nn.state import safe_load, safe_save
-from tinygrad.llm.model import Transformer, TransformerConfig, SSMConfig
-from tinygrad.llm.quant import quantize_mxfp4_cpu
-
-KIMI_SSM_LAYERS = tuple(i not in (3, 7, 11, 15, 19, 23, 26) for i in range(27))
-KIMI_TENSOR_COUNT, KIMI_LOGICAL_BYTES = 688, 29_051_930_368
-KIMI_CHECKPOINT_FORMAT = "tinygrad-kimi-mxfp4-v2"
-
-def kimi_config(max_context:int, expert_mxfp4:bool=True) -> TransformerConfig:
-  return TransformerConfig(num_blocks=27, dim=2304, hidden_dim=1024, n_heads=32, n_kv_heads=32, norm_eps=1e-5,
-    vocab_size=163840, head_dim=192, rope_theta=10000.0, rope_dim=64, v_head_dim=128, q_lora_rank=0, kv_lora_rank=512,
-    num_experts=256, num_experts_per_tok=8, norm_topk_prob=True, shared_expert_dim=1024, leading_dense_blocks=1,
-    dense_hidden_dim=9216, routed_scaling_factor=2.446, expert_bias=True, max_context=max_context, expert_mxfp4=expert_mxfp4,
-    shared_expert_gate=False, bf16_activations=True, kda_split_qkv=True,
-    recurrent_prefill_chunked=True, recurrent_prefill_chunk_size=32,
-    ssm=SSMConfig(conv_kernel=4, state_size=128, group_count=32, time_step_rank=32, inner_size=4096, kda=True),
-    ssm_layers=KIMI_SSM_LAYERS)
-
-def _shard_kimi(model:Transformer, devices:tuple[str, ...]) -> None:
-  """Tensor-parallel layout. Every GPU owns a slice of every expert (not a replicated expert set)."""
-  for name, value in nn.state.get_state_dict(model).items():
-    axis = None
-    if name in ("token_embd.weight", "output.weight"): axis = 0
-    elif ".ffn_gate_exps.weight" in name or ".ffn_up_exps.weight" in name: axis = 1
-    elif ".ffn_gate_exps.weight_scale" in name or ".ffn_up_exps.weight_scale" in name: axis = 1
-    elif ".ffn_down_exps.weight" in name or ".ffn_down_exps.weight_scale" in name: axis = 2
-    elif name.endswith((".ffn_gate.weight", ".ffn_up.weight", ".ffn_gate_shexp.weight", ".ffn_up_shexp.weight")): axis = 0
-    elif name.endswith((".ffn_down.weight", ".ffn_down_shexp.weight", ".attn_output.weight", ".ssm_out.weight")): axis = 1
-    elif name.endswith((".attn_q.weight", ".attn_k.weight", ".attn_v.weight", ".attn_qkv.weight",
-                        ".ssm_f_b.weight", ".ssm_g_b.weight", ".ssm_beta.weight")): axis = 0
-    elif name.endswith((".ssm_conv1d.weight", ".ssm_q_conv1d.weight", ".ssm_k_conv1d.weight", ".ssm_v_conv1d.weight")): axis = 0
-    elif name.endswith((".ssm_a", ".ssm_dt.bias")): axis = 0
-    elif name.endswith((".attn_k_b.weight", ".attn_v_b.weight")): axis = 0
-    value.shard_(devices, axis=axis)
-
-def _validate_kimi_state(model:Transformer, state:dict[str, Tensor]) -> None:
-  model_state = nn.state.get_state_dict(model)
-  missing, unexpected = set(model_state)-set(state), set(state)-set(model_state)
-  if missing or unexpected: raise ValueError(f"invalid Kimi tensor names: missing={sorted(missing)}, unexpected={sorted(unexpected)}")
-  for name, value in state.items():
-    if value.shape != model_state[name].shape: raise ValueError(f"invalid shape for {name}: expected {model_state[name].shape}, got {value.shape}")
-    expected_dtype = dtypes.uint8 if name.endswith((".weight_scale", "_exps.weight")) else dtypes.bfloat16
-    if value.dtype != expected_dtype: raise ValueError(f"invalid dtype for {name}: expected {expected_dtype}, got {value.dtype}")
-  if len(state) != KIMI_TENSOR_COUNT or (nbytes := sum(x.nbytes() for x in state.values())) != KIMI_LOGICAL_BYTES:
-    raise ValueError(f"invalid Kimi checkpoint size: {len(state)} tensors, {nbytes} bytes")
-
-def _load_converted_state(model_dir:pathlib.Path, files:list[str]) -> dict[str, Tensor]:
-  state:dict[str, Tensor] = {}
-  for filename in files:
-    part = safe_load(model_dir / filename)
-    if duplicates := set(state) & set(part): raise ValueError(f"duplicate Kimi tensors in {filename}: {sorted(duplicates)}")
-    state.update(part)
-  return state
-
-def load_kimi(model_dir:str|pathlib.Path, max_context:int=4096, devices:int=4) -> Transformer:
-  model_dir = pathlib.Path(model_dir)
-  manifest = json.loads((model_dir / "tinygrad-kimi.json").read_text())
-  if manifest.get("format") != KIMI_CHECKPOINT_FORMAT: raise ValueError("unsupported Kimi checkpoint format")
-  if devices != 4: raise ValueError("Kimi-Linear MXFP4 checkpoint currently requires TP4 (--devices 4)")
-  devs = tuple(f"{Device.DEFAULT}:{i}" for i in range(devices))
-  model = Transformer(kimi_config(max_context, expert_mxfp4=True))
-  state = _load_converted_state(model_dir, manifest["files"])
-  _validate_kimi_state(model, state)
-  _shard_kimi(model, devs)
-  nn.state.load_state_dict(model, state, strict=True, consume=True, realize=True)
-  if state: raise ValueError(f"unexpected Kimi tensors: {sorted(state)}")
-  return model
-
-def _load_hf_state(src:pathlib.Path) -> dict[str, Tensor]:
-  index = json.loads((src / "model.safetensors.index.json").read_text())
-  state:dict[str, Tensor] = {}
-  for filename in sorted(set(index["weight_map"].values())): state.update(safe_load(src / filename))
-  return state
-
-def _layer_key(i:int, suffix:str) -> str: return f"model.layers.{i}.{suffix}"
-
-def _convert_attention(sd:dict[str, Tensor], i:int, is_kda:bool, consume:bool=False) -> dict[str, Tensor]:
-  p, out = f"blk.{i}.", {}
-  def get(suffix:str) -> Tensor:
-    key = _layer_key(i, suffix)
-    return (sd.pop(key) if consume else sd[key]).to("CPU")
-  if is_kda:
-    for src_name, dst_name in (("q_proj", "attn_q"), ("k_proj", "attn_k"), ("v_proj", "attn_v")):
-      out[p+dst_name+".weight"] = get(f"self_attn.{src_name}.weight")
-    for src_name, dst_name in (("q_conv1d", "ssm_q_conv1d"), ("k_conv1d", "ssm_k_conv1d"), ("v_conv1d", "ssm_v_conv1d")):
-      out[p+dst_name+".weight"] = get(f"self_attn.{src_name}.weight").squeeze(1)
-    for src_name, dst_name in (("f_a_proj", "ssm_f_a"), ("f_b_proj", "ssm_f_b"), ("g_a_proj", "ssm_g_a"),
-                               ("g_b_proj", "ssm_g_b"), ("b_proj", "ssm_beta"), ("o_proj", "ssm_out")):
-      out[p+dst_name+".weight"] = get(f"self_attn.{src_name}.weight")
-    out[p+"ssm_norm.weight"] = get("self_attn.o_norm.weight")
-    out[p+"ssm_dt.bias"] = get("self_attn.dt_bias")
-    out[p+"ssm_a"] = (-get("self_attn.A_log").float().exp()).reshape(32, 1)
-  else:
-    out[p+"attn_q.weight"] = get("self_attn.q_proj.weight")
-    out[p+"attn_kv_a_mqa.weight"] = get("self_attn.kv_a_proj_with_mqa.weight")
-    out[p+"attn_kv_a_norm.weight"] = get("self_attn.kv_a_layernorm.weight")
-    kv_b = get("self_attn.kv_b_proj.weight").reshape(32, 256, 512)
-    k_b, v_b = kv_b[:, :128], kv_b[:, 128:]
-    out[p+"attn_k_b.weight"], out[p+"attn_v_b.weight"] = k_b.transpose(1, 2), v_b
-    out[p+"attn_output.weight"] = get("self_attn.o_proj.weight")
-  return out
-
-def convert_kimi(src_dir:str|pathlib.Path, dst_dir:str|pathlib.Path) -> None:
-  """Stream the official BF16 checkpoint into the tinygrad TP4 MXFP4/BF16 representation."""
-  src, dst = pathlib.Path(src_dir), pathlib.Path(dst_dir)
-  dst.mkdir(parents=True, exist_ok=True)
-  config = json.loads((src / "config.json").read_text())
-  expected = {"hidden_size":2304, "num_hidden_layers":27, "num_attention_heads":32, "num_key_value_heads":32,
-              "vocab_size":163840, "intermediate_size":9216, "num_experts":256, "num_experts_per_token":8,
-              "moe_intermediate_size":1024, "num_shared_experts":1, "qk_nope_head_dim":128, "qk_rope_head_dim":64,
-              "v_head_dim":128, "kv_lora_rank":512, "first_k_dense_replace":1, "mla_use_nope":True}
-  if any(config.get(k) != v for k,v in expected.items()): raise ValueError(f"not Kimi-Linear-48B-A3B: expected {expected}")
-  sd, files = _load_hf_state(src), []
-
-  common = {"token_embd.weight":sd.pop("model.embed_tokens.weight").to("CPU"), "output_norm.weight":sd.pop("model.norm.weight").to("CPU"),
-            "output.weight":sd.pop("lm_head.weight").to("CPU")}
-  safe_save(common, str(dst / "model-common.safetensors"))
-  del common
-  gc.collect()
-  files.append("model-common.safetensors")
-  for i in range(27):
-    p = f"blk.{i}."
-    layer = _convert_attention(sd, i, KIMI_SSM_LAYERS[i], consume=True)
-    layer[p+"attn_norm.weight"] = sd.pop(_layer_key(i, "input_layernorm.weight")).to("CPU")
-    layer[p+"ffn_norm.weight"] = sd.pop(_layer_key(i, "post_attention_layernorm.weight")).to("CPU")
-    if i == 0:
-      for src_name, dst_name in (("gate_proj", "ffn_gate"), ("up_proj", "ffn_up"), ("down_proj", "ffn_down")):
-        layer[p+dst_name+".weight"] = sd.pop(_layer_key(i, f"mlp.{src_name}.weight")).to("CPU")
-    else:
-      base = _layer_key(i, "block_sparse_moe")
-      layer[p+"ffn_gate_inp.weight"] = sd.pop(base+".gate.weight").to("CPU")
-      # The official name is e_score_correction_bias; tolerate the early checkpoint spelling.
-      bias_name = next(k for k in (base+".gate.e_score_correction_bias", base+".gate.e_score_correction") if k in sd)
-      layer[p+"exp_probs_b.bias"] = sd.pop(bias_name).to("CPU")
-      for src_name, dst_name in (("gate_proj", "ffn_gate_shexp"), ("up_proj", "ffn_up_shexp"), ("down_proj", "ffn_down_shexp")):
-        layer[p+dst_name+".weight"] = sd.pop(base+f".shared_experts.{src_name}.weight").to("CPU")
-    layer_file = f"model-layer-{i:02d}.safetensors"
-    safe_save({k:v.cast(dtypes.bfloat16).contiguous() for k,v in layer.items()}, str(dst/layer_file))
-    del layer
-    gc.collect()
-    files.append(layer_file)
-
-    if i:
-      base = _layer_key(i, "block_sparse_moe.experts")
-      for wid, dst_name in (("w1", "ffn_gate_exps"), ("w3", "ffn_up_exps"), ("w2", "ffn_down_exps")):
-        packed, scales = [], []
-        for e in range(256):
-          q, s = quantize_mxfp4_cpu(sd.pop(f"{base}.{e}.{wid}.weight").to("CPU"))
-          packed.append(q)
-          scales.append(s)
-        expert_file = f"model-layer-{i:02d}-{wid}-mxfp4.safetensors"
-        safe_save({p+dst_name+".weight":Tensor.stack(*packed), p+dst_name+".weight_scale":Tensor.stack(*scales)}, str(dst/expert_file))
-        del packed, scales, q, s
-        gc.collect()
-        files.append(expert_file)
-
-  if sd: raise ValueError(f"unconverted Kimi source tensors: {sorted(sd)}")
-  for name in ("config.json", "tokenizer_config.json", "special_tokens_map.json", "tiktoken.model", "chat_template.jinja"):
-    if (src/name).exists(): shutil.copy2(src/name, dst/name)
-  converted = _load_converted_state(dst, files)
-  _validate_kimi_state(Transformer(kimi_config(max_context=1)), converted)
-  manifest = {"format":KIMI_CHECKPOINT_FORMAT, "tensor_count":KIMI_TENSOR_COUNT,
-              "logical_bytes":KIMI_LOGICAL_BYTES, "files":files}
-  (dst / "tinygrad-kimi.json").write_text(json.dumps(manifest, indent=2)+"\n")
-
-def load_kimi_tokenizer_data(model_dir:str|pathlib.Path) -> tuple[dict[str, int], dict[str, int], int, int]:
-  """Return byte-encoded normal tokens and specials for SimpleTokenizer without transformers/tiktoken."""
-  model_dir = pathlib.Path(model_dir)
-  normal:dict[str, int] = {}
-  bs = [*range(33, 127), *range(161, 173), *range(174, 256)]
-  byte_encoder = {b:chr(b) for b in bs} | {b:chr(256+i) for i,b in enumerate(b for b in range(256) if b not in bs)}
-  for line in (model_dir / "tiktoken.model").read_bytes().splitlines():
-    token, rank = line.split()
-    normal["".join(byte_encoder[b] for b in base64.b64decode(token))] = int(rank)
-  tc = json.loads((model_dir / "tokenizer_config.json").read_text())
-  specials = {v["content"]:int(k) for k,v in tc.get("added_tokens_decoder", {}).items()}
-  cfg = json.loads((model_dir / "config.json").read_text())
-  return normal, specials, cfg["bos_token_id"], cfg["eos_token_id"]

tinygrad/llm/kimi_k3.py

@@ -1,337 +0,0 @@
-from __future__ import annotations
-import gc, json, math, pathlib
-from dataclasses import replace
-from collections import defaultdict
-from typing import Callable, cast
-from tinygrad import Tensor, Device, dtypes, nn
-from tinygrad.device import Buffer
-from tinygrad.uop.ops import UOp
-from tinygrad.nn.state import safe_dtypes, safe_load_metadata
-from tinygrad.llm.kimi import load_kimi_tokenizer_data
-from tinygrad.llm.model import SSMConfig, Transformer, TransformerConfig
-
-KIMI_K3_TOTAL_SIZE = 1_560_860_324_864
-KIMI_K3_TEXT_SIZE = 1_559_965_606_912
-KIMI_K3_TP8_BYTES_PER_GPU = 196_784_397_312
-KIMI_K3_SHARDS = 96
-KIMI_K3_EXPERTS = 896
-KIMI_K3_LAYERS = 93
-KIMI_K3_FULL_ATTN_LAYERS = (*range(3, KIMI_K3_LAYERS, 4), 92)
-KIMI_K3_SSM_LAYERS = tuple(i not in KIMI_K3_FULL_ATTN_LAYERS for i in range(KIMI_K3_LAYERS))
-
-def kimi_k3_config(max_context:int) -> TransformerConfig:
-  """Official Kimi K3 text-tower configuration (zero-based full-attention layers)."""
-  return TransformerConfig(num_blocks=93, dim=7168, hidden_dim=3072, n_heads=96, n_kv_heads=96, norm_eps=1e-5,
-    vocab_size=163840, head_dim=192, rope_theta=10000.0, rope_dim=64, v_head_dim=128, max_context=max_context,
-    q_lora_rank=1536, kv_lora_rank=512, num_experts=896, num_experts_per_tok=16, norm_topk_prob=True,
-    shared_expert_dim=6144, leading_dense_blocks=1, dense_hidden_dim=33792, routed_scaling_factor=1.0,
-    expert_bias=True, expert_mxfp4=True, bf16_activations=True, kda_split_qkv=True,
-    ssm=SSMConfig(conv_kernel=4, state_size=128, group_count=96, time_step_rank=96, inner_size=12288, kda=True, channel_decay=True),
-    ssm_layers=KIMI_K3_SSM_LAYERS, shared_expert_gate=False, attn_output_gate=True,
-    activation_situ_beta=4.0, activation_situ_linear_beta=25.0, routed_expert_dim=3584, latent_moe_norm=True,
-    route_weights_uncorrected=True, attn_res_block_size=12, kda_full_rank_gate=True, kda_gate_lower_bound=-5.0,
-    recurrent_prefill_chunked=True, recurrent_prefill_chunk_size=128)
-
-def kimi_k3_smoke_config(max_context:int=4) -> TransformerConfig:
-  """Reduced K3 with every architectural feature retained for cheap compile/hardware admission tests."""
-  # Keep both the routed latent and the TP8-local expert hidden dimension wave64 aligned. The real
-  # gfx950 packed-expert kernel requires this, so the hardware smoke test must preserve the constraint.
-  return replace(kimi_k3_config(max_context), num_blocks=2, dim=32, hidden_dim=512, n_heads=8, n_kv_heads=8,
-    vocab_size=64, head_dim=8, rope_dim=4, v_head_dim=4, q_lora_rank=16, kv_lora_rank=8, num_experts=512,
-    num_experts_per_tok=2, shared_expert_dim=32, dense_hidden_dim=64, routed_expert_dim=64,
-    ssm=SSMConfig(4, 4, 8, 8, 32, True, True), ssm_layers=(True, False), attn_res_block_size=1)
-
-def _shard_kimi_k3(model:Transformer, devices:tuple[str, ...]) -> None:
-  """Tensor parallel layout for K3. The official dimensions are divisible by TP8."""
-  if len(devices) not in (1, 2, 4, 8): raise ValueError(f"Kimi K3 tensor parallelism requires 1, 2, 4, or 8 devices, got {len(devices)}")
-  for name, value in nn.state.get_state_dict(model).items():
-    axis = None
-    if name in ("token_embd.weight", "output.weight"): axis = 0
-    elif ".ffn_gate_exps.weight" in name or ".ffn_up_exps.weight" in name: axis = 1
-    elif ".ffn_gate_exps.weight_scale" in name or ".ffn_up_exps.weight_scale" in name: axis = 1
-    elif ".ffn_down_exps.weight" in name or ".ffn_down_exps.weight_scale" in name: axis = 2
-    elif name.endswith((".ffn_gate.weight", ".ffn_up.weight", ".ffn_gate_shexp.weight", ".ffn_up_shexp.weight")): axis = 0
-    elif name.endswith((".ffn_down.weight", ".ffn_down_shexp.weight", ".ffn_routed_down.weight", ".ffn_routed_up.weight",
-                        ".attn_output.weight", ".ssm_out.weight")): axis = 1
-    elif name.endswith((".attn_q_b.weight", ".attn_k_b.weight", ".attn_v_b.weight", ".attn_gate.weight",
-                        ".attn_q.weight", ".attn_k.weight", ".attn_v.weight", ".ssm_f_b.weight", ".ssm_g_full.weight", ".ssm_beta.weight")): axis = 0
-    elif name.endswith((".ssm_q_conv1d.weight", ".ssm_k_conv1d.weight", ".ssm_v_conv1d.weight", ".ssm_dt.bias")): axis = 0
-    value.shard_(devices, axis=axis)
-
-def _validate_config(config:dict) -> None:
-  text = config.get("text_config", config)
-  expected = {"model_type":"kimi_linear", "hidden_size":7168, "num_hidden_layers":93, "num_attention_heads":96,
-              "vocab_size":163840, "intermediate_size":33792, "num_experts":896, "num_experts_per_token":16,
-              "moe_intermediate_size":3072, "num_shared_experts":2, "q_lora_rank":1536, "kv_lora_rank":512,
-              "qk_nope_head_dim":128, "qk_rope_head_dim":64, "v_head_dim":128, "routed_expert_hidden_size":3584,
-              "attn_res_block_size":12, "hidden_act":"situ", "mla_use_nope":True, "mla_use_output_gate":True,
-              "activation_situ_beta":4.0, "activation_situ_linear_beta":25.0, "latent_moe_use_norm":True,
-              "moe_renormalize":True, "first_k_dense_replace":1, "num_expert_group":1, "topk_group":1}
-  bad = {k:(text.get(k), v) for k,v in expected.items() if text.get(k) != v}
-  linear = text.get("linear_attn_config", {})
-  linear_expected = {"head_dim":128, "num_heads":96, "short_conv_kernel_size":4, "use_full_rank_gate":True,
-                     "gate_lower_bound":-5.0, "full_attn_layers":[i+1 for i in KIMI_K3_FULL_ATTN_LAYERS],
-                     "kda_layers":[i+1 for i,x in enumerate(KIMI_K3_SSM_LAYERS) if x]}
-  bad.update({f"linear_attn_config.{k}":(linear.get(k), v) for k,v in linear_expected.items() if linear.get(k) != v})
-  quant = text.get("quantization_config", {})
-  if quant.get("format") != "mxfp4-pack-quantized": bad["quantization_config.format"] = (quant.get("format"), "mxfp4-pack-quantized")
-  if bad: raise ValueError(f"not the supported official Kimi K3 checkpoint: {bad}")
-
-def audit_kimi_k3_checkpoint(model_dir:str|pathlib.Path, require_shards:bool=True) -> dict[str, int]:
-  """Validate checkpoint metadata only. This never opens weight data and is safe on small hosts."""
-  root = pathlib.Path(model_dir)
-  _validate_config(json.loads((root / "config.json").read_text()))
-  index = json.loads((root / "model.safetensors.index.json").read_text())
-  weight_map, total = index.get("weight_map", {}), index.get("metadata", {}).get("total_size")
-  language = [k for k in weight_map if k.startswith("language_model.")]
-  experts = [k for k in language if ".block_sparse_moe.experts." in k]
-  missing_files = {fn for fn in weight_map.values() if not (root / fn).is_file()}
-  if total != KIMI_K3_TOTAL_SIZE: raise ValueError(f"unexpected checkpoint size {total}, expected {KIMI_K3_TOTAL_SIZE}")
-  if len(set(weight_map.values())) != KIMI_K3_SHARDS: raise ValueError("official Kimi K3 must contain 96 safetensor shards")
-  if len(experts) != 92 * KIMI_K3_EXPERTS * 3 * 2: raise ValueError(f"unexpected routed-expert tensor count {len(experts)}")
-  if require_shards and missing_files: raise FileNotFoundError(f"missing {len(missing_files)} checkpoint shards, first: {sorted(missing_files)[0]}")
-  return {"tensors":len(weight_map), "language_tensors":len(language), "expert_tensors":len(experts),
-          "shards":len(set(weight_map.values())), "missing_shards":len(missing_files), "total_size":total}
-
-def _layer_sources(i:int, is_kda:bool) -> dict[str, str]:
-  src, dst = f"language_model.model.layers.{i}.", f"blk.{i}."
-  out = {
-    src+"input_layernorm.weight":dst+"attn_norm.weight", src+"post_attention_layernorm.weight":dst+"ffn_norm.weight",
-    src+"self_attention_res_norm.weight":dst+"attn_res_norm.weight", src+"self_attention_res_proj.weight":dst+"attn_res_proj.weight",
-    src+"mlp_res_norm.weight":dst+"mlp_res_norm.weight", src+"mlp_res_proj.weight":dst+"mlp_res_proj.weight",
-  }
-  if is_kda:
-    for a,b in (("q_proj","attn_q"),("k_proj","attn_k"),("v_proj","attn_v"),("g_proj","ssm_g_full"),
-                ("f_a_proj","ssm_f_a"),("f_b_proj","ssm_f_b"),("b_proj","ssm_beta"),("o_proj","ssm_out")):
-      out[src+f"self_attn.{a}.weight"] = dst+b+".weight"
-    for a,b in (("q_conv1d","ssm_q_conv1d"),("k_conv1d","ssm_k_conv1d"),("v_conv1d","ssm_v_conv1d")):
-      out[src+f"self_attn.{a}.weight"] = dst+b+".weight"
-    out[src+"self_attn.o_norm.weight"], out[src+"self_attn.dt_bias"], out[src+"self_attn.A_log"] = \
-      dst+"ssm_norm.weight", dst+"ssm_dt.bias", dst+"ssm_a"
-  else:
-    for a,b in (("q_a_proj","attn_q_a"),("q_a_layernorm","attn_q_a_norm"),("q_b_proj","attn_q_b"),
-                ("kv_a_proj_with_mqa","attn_kv_a_mqa"),("kv_a_layernorm","attn_kv_a_norm"),
-                ("g_proj","attn_gate"),("o_proj","attn_output")):
-      out[src+f"self_attn.{a}.weight"] = dst+b+".weight"
-    # kv_b_proj is split into head-wise K and V tensors while loading.
-    out[src+"self_attn.kv_b_proj.weight"] = dst+"attn_k_b.weight|"+dst+"attn_v_b.weight"
-  if i == 0:
-    for a,b in (("gate_proj","ffn_gate"),("up_proj","ffn_up"),("down_proj","ffn_down")): out[src+f"mlp.{a}.weight"] = dst+b+".weight"
-  else:
-    base = src+"block_sparse_moe."
-    out[base+"gate.weight"], out[base+"gate.e_score_correction_bias"] = dst+"ffn_gate_inp.weight", dst+"exp_probs_b.bias"
-    for a,b in (("gate_proj","ffn_gate_shexp"),("up_proj","ffn_up_shexp"),("down_proj","ffn_down_shexp"),
-                ("routed_expert_down_proj","ffn_routed_down"),("routed_expert_up_proj","ffn_routed_up"),
-                ("routed_expert_norm","ffn_routed_norm")):
-      out[base+(f"shared_experts.{a}.weight" if a.endswith("_proj") and not a.startswith("routed_") else a+".weight")] = dst+b+".weight"
-  return out
-
-def _replace(dst:Tensor, src:Tensor) -> None:
-  if dst.shape != src.shape: raise ValueError(f"shape mismatch: expected {dst.shape}, got {src.shape}")
-  if not isinstance(dst.device, tuple):
-    dst.replace(src.to(dst.device)).realize()
-    return
-  if isinstance(src.device, tuple):
-    dst.replace(src.shard_like(dst)).realize()
-    return
-
-  # Build the final MultiBuffer directly. The generic shard().realize() path schedules several
-  # kernels per tensor and recompiles them for every DISK:<filename> device. Axis-0 shards and
-  # replicas are contiguous, so copy those bytes straight into their final device buffers.
-  devices, axis, shape = dst.device, dst.uop.axis, tuple(int(x) for x in dst.shape)
-  try: src_buffer = cast(Buffer, src.uop.buffer)
-  except (AssertionError, RuntimeError):
-    src = src.clone().realize()
-    src_buffer = cast(Buffer, src.uop.buffer)
-
-  if axis is None:
-    # Replicas are identical on every device. Read the disk tensor once, retain that allocation on
-    # GPU 0, and fan it out over XGMI instead of issuing eight identical direct reads.
-    staging = Tensor.empty(*shape, dtype=src.dtype, device=devices[0]).realize()
-    cast(Buffer, staging.uop.buffer).ensure_allocated().copy_from(src_buffer.ensure_allocated())
-    parts = [staging]
-    for device in devices[1:]:
-      part = Tensor.empty(*shape, dtype=src.dtype, device=device).realize()
-      cast(Buffer, part.uop.buffer).ensure_allocated().copy_from(cast(Buffer, staging.uop.buffer).ensure_allocated())
-      parts.append(part)
-    dst.replace(Tensor(parts[0].uop.mstack(*(x.uop for x in parts[1:]))))
-    return
-  if axis == 0:
-    part_shape = (shape[0]//len(devices), *shape[1:])
-    part_numel = math.prod(part_shape)
-    parts:list[Tensor] = []
-    for i,device in enumerate(devices):
-      part = Tensor.empty(*part_shape, dtype=src.dtype, device=device).realize()
-      source = src_buffer.view(part_numel, src.dtype, i*part_numel*src.dtype.itemsize)
-      cast(Buffer, part.uop.buffer).ensure_allocated().copy_from(source.ensure_allocated())
-      parts.append(part)
-  else:
-    # Inner-axis TP slices are strided in row-major safetensors. Stage one complete tensor on
-    # GPU 0, then schedule all slice kernels and peer copies as one multi-device graph.
-    staging = Tensor.empty(*shape, dtype=src.dtype, device=devices[0]).realize()
-    cast(Buffer, staging.uop.buffer).ensure_allocated().copy_from(src_buffer.ensure_allocated())
-    dst.replace(staging.shard(devices, axis=axis)).realize()
-    return
-  dst.replace(Tensor(parts[0].uop.mstack(*(x.uop for x in parts[1:])).unshard(axis)))
-
-def _safe_load_selected(fn:pathlib.Path, keys:tuple[str, ...]|list[str]) -> dict[str, Tensor]:
-  """Create disk-backed tensors only for selected safetensor entries, without touching payload data."""
-  source, data_start, metadata = safe_load_metadata(fn)
-  data = source[data_start:]
-  missing = [key for key in keys if key not in metadata]
-  if missing: raise ValueError(f"missing tensor {missing[0]} from {fn.name}")
-  out:dict[str, Tensor] = {}
-  for key in keys:
-    entry = metadata[key]
-    out[key] = data[entry["data_offsets"][0]:entry["data_offsets"][1]].bitcast(safe_dtypes[entry["dtype"]]).reshape(entry["shape"])
-  return out
-
-def _load_stacked_experts(dst:Tensor, sources:list[Tensor]) -> None:
-  """Read expert tensors once into a transient GPU staging buffer, then redistribute TP slices over the GPU fabric."""
-  if not sources or not isinstance(dst.device, tuple) or dst.uop.axis is None: raise ValueError("expected a TP-sharded expert destination")
-  devices, axis = dst.device, dst.uop.axis
-  shape = (len(sources), *sources[0].shape)
-  if dst.shape != shape or any(x.shape != sources[0].shape or x.dtype != dst.dtype for x in sources):
-    raise ValueError(f"expert source shape/dtype does not match destination {dst.shape} {dst.dtype}")
-
-  # Each official expert tensor is contiguous in its safetensor file. Assemble it once on GPU 0;
-  # Buffer.copy_from uses the AMD driver's bounded DISK->GPU staging path and never allocates host-sized storage.
-  staging = Tensor.empty(*shape, dtype=dst.dtype, device=devices[0]).realize()
-  staging_buffer = cast(Buffer, staging.uop.buffer)
-  def free_staging_cache() -> None:
-    if (free_cache:=getattr(Device[devices[0]].allocator, "free_cache", None)) is not None: free_cache()
-  offset = 0
-  for source in sources:
-    source_buffer = cast(Buffer, source.uop.buffer)
-    staging_buffer.view(cast(int, source.numel()), source.dtype, offset).ensure_allocated().copy_from(source_buffer.ensure_allocated())
-    offset += source.nbytes()
-
-  # Schedule all TP slices and peer copies together. This avoids eight independent realization
-  # passes and lets the runtime overlap the multi-device transfer graph.
-  dst.replace(staging.shard(devices, axis=axis)).realize()
-  del staging
-  gc.collect()
-  free_staging_cache()
-
-def _load_nonexperts(root:pathlib.Path, weight_map:dict[str, str], model:Transformer, progress:Callable[[str], None]) -> set[str]:
-  model_state, mappings = nn.state.get_state_dict(model), {
-    "language_model.model.embed_tokens.weight":"token_embd.weight", "language_model.model.norm.weight":"output_norm.weight",
-    "language_model.lm_head.weight":"output.weight", "language_model.model.output_attn_res_norm.weight":"output_attn_res_norm.weight",
-    "language_model.model.output_attn_res_proj.weight":"output_attn_res_proj.weight"}
-  for i,is_kda in enumerate(KIMI_K3_SSM_LAYERS): mappings.update(_layer_sources(i, is_kda))
-  by_file:dict[str, list[str]] = defaultdict(list)
-  for source in mappings:
-    if source not in weight_map: raise ValueError(f"missing Kimi K3 tensor {source}")
-    by_file[weight_map[source]].append(source)
-  consumed:set[str] = set()
-  for filename, sources in sorted(by_file.items()):
-    progress(f"loading non-expert tensors from {filename}")
-    shard = _safe_load_selected(root / filename, sources)
-    for source in sources:
-      value, targets = shard[source], mappings[source].split("|")
-      # A_log is the only checkpoint tensor requiring arithmetic during load. Realize its 128
-      # channel values on CPU so replicating it does not try to render the disk/PYTHON graph.
-      if source.endswith("A_log"): value = (-value.to("CPU").float().exp()).reshape(model_state[targets[0]].shape).realize()
-      if source.endswith("conv1d.weight"): value = value.squeeze(1)
-      if source.endswith("kv_b_proj.weight"):
-        # Splitting K/V includes a transpose, which cannot be rendered against a disk buffer.
-        # Materialize only this one 25 MiB projection on CPU, then release it with the shard.
-        value = value.to("CPU").realize().reshape(96, 256, 512)
-        values:tuple[Tensor, ...] = (value[:, :128].transpose(1, 2), value[:, 128:])
-      else: values = (value,)
-      for target,tensor in zip(targets, values): _replace(model_state[target], tensor)
-      consumed.add(source)
-    del shard
-  gc.collect()
-  return consumed
-
-def _load_experts(root:pathlib.Path, weight_map:dict[str, str], model:Transformer, progress:Callable[[str], None]) -> set[str]:
-  model_state, consumed = nn.state.get_state_dict(model), set[str]()
-  for i in range(1, KIMI_K3_LAYERS):
-    base = f"language_model.model.layers.{i}.block_sparse_moe.experts"
-    fields = tuple((wid, suffix, dst_name) for wid,dst_name in (("w1","ffn_gate_exps"),("w2","ffn_down_exps"),("w3","ffn_up_exps"))
-                   for suffix in ("weight_packed", "weight_scale"))
-    keys = {(e,wid,suffix):f"{base}.{e}.{wid}.{suffix}" for e in range(KIMI_K3_EXPERTS) for wid,suffix,_ in fields}
-    files = sorted({weight_map[k] for k in keys.values()})
-    progress(f"loading layer {i}/92 routed experts from {', '.join(files)}")
-    shards = {fn:_safe_load_selected(root / fn, [key for key in keys.values() if weight_map[key] == fn]) for fn in files}
-
-    # Official K3 stores all six tensors for an expert contiguously and all 896 experts for a
-    # layer in one contiguous shard region, ordered lexicographically by expert name. Read that
-    # region once, then reorder/split on GPU 0 directly into the six TP8 destinations.
-    blocks:list[tuple[int, int, int, list[Buffer]]] = []
-    for e in range(KIMI_K3_EXPERTS):
-      bufs = [cast(Buffer, shards[weight_map[keys[e,wid,suffix]]][keys[e,wid,suffix]].uop.buffer) for wid,suffix,_ in fields]
-      if not all(bufs[j].device == bufs[0].device and bufs[j].offset+bufs[j].nbytes == bufs[j+1].offset for j in range(len(bufs)-1)):
-        raise ValueError(f"layer {i} expert {e} tensors are not contiguous in the official shard")
-      blocks.append((bufs[0].offset, bufs[-1].offset+bufs[-1].nbytes, e, bufs))
-    blocks.sort()
-    if len(files) != 1 or not all(blocks[j][1] == blocks[j+1][0] for j in range(len(blocks)-1)):
-      raise ValueError(f"layer {i} routed experts are not one contiguous official-shard region")
-    row_bytes = blocks[0][1]-blocks[0][0]
-    if any(end-start != row_bytes for start,end,_,_ in blocks): raise ValueError(f"layer {i} expert records have inconsistent sizes")
-
-    devices = cast(tuple[str, ...], model_state[f"blk.{i}.ffn_gate_exps.weight"].device)
-    raw = Tensor.empty(KIMI_K3_EXPERTS, row_bytes, dtype=dtypes.uint8, device=devices[0]).realize()
-    raw_buffer, first_buffer = cast(Buffer, raw.uop.buffer), blocks[0][3][0]
-    raw_buffer.ensure_allocated().copy_from(first_buffer.base.view(KIMI_K3_EXPERTS*row_bytes, dtypes.uint8, blocks[0][0]).ensure_allocated())
-    lexpos = {expert:pos for pos,(_,_,expert,_) in enumerate(blocks)}
-    permutation = Tensor([lexpos[e] for e in range(KIMI_K3_EXPERTS)], device=devices[0])
-    field_offset, outputs = 0, []
-    for field_idx,(wid,suffix,dst_name) in enumerate(fields):
-      field_bytes = blocks[0][3][field_idx].nbytes
-      dst = model_state[f"blk.{i}.{dst_name}.weight" + ("_scale" if suffix == "weight_scale" else "")]
-      axis = dst.uop.axis
-      if axis is None: raise ValueError(f"layer {i} expert destination {dst_name} is not TP-sharded")
-      value = raw[:, field_offset:field_offset+field_bytes][permutation].reshape(dst.shape).shard(devices, axis=axis)
-      # Realize into a buffer-identity tensor, then retain only that identity in the model. Keeping
-      # value's arithmetic UOp would also keep the 15.7 GB raw staging tensor and its reorder graph
-      # alive for every loaded weight, wasting about 44 GB on GPU 0 after the load completes.
-      shard_shape = tuple(int(x) for x in value.uop.shard_shape)
-      storage = UOp.new_buffer(devices, math.prod(shard_shape), dst.dtype).reshape(shard_shape).unshard(axis)
-      final = Tensor(storage)
-      final.assign(value)
-      outputs.append((dst, final, storage))
-      field_offset += field_bytes
-    if field_offset != row_bytes: raise ValueError(f"layer {i} expert field sizes do not cover the contiguous record")
-    outputs[0][1].realize(*(value for _,value,_ in outputs[1:]))
-    for dst,_,storage in outputs: dst.replace(Tensor(storage))
-    consumed.update(keys.values())
-    # Drop the realized assignment graphs before flushing the allocator cache. Their final storage
-    # UOps remain in model_state, while the graphs themselves still reference raw and permutation.
-    del shards, raw, raw_buffer, permutation, outputs, value, final, storage, dst
-    if (free_cache:=getattr(Device[devices[0]].allocator, "free_cache", None)) is not None: free_cache()
-  return consumed
-
-def load_kimi_k3(model_dir:str|pathlib.Path, max_context:int=4096, devices:int=8,
-                 progress:Callable[[str], None]=print) -> Transformer:
-  """Load the official native K3 checkpoint without ever materializing it in host RAM.
-
-  Safetensor shards remain disk-backed. Expert tensors are read once into a bounded GPU staging
-  buffer, redistributed as TP slices, and discarded after every projection. Vision tensors are intentionally ignored.
-  """
-  root = pathlib.Path(model_dir)
-  _validate_config(json.loads((root / "config.json").read_text()))
-  index = json.loads((root / "model.safetensors.index.json").read_text())
-  weight_map = index["weight_map"]
-  if devices != 8: raise ValueError("official Kimi K3 currently requires --devices 8")
-  if index.get("metadata", {}).get("total_size") != KIMI_K3_TOTAL_SIZE or len(set(weight_map.values())) != KIMI_K3_SHARDS:
-    raise ValueError("checkpoint index does not match the official 96-shard Kimi K3 release")
-  missing_files = {fn for fn in weight_map.values() if not (root / fn).is_file()}
-  if missing_files: raise FileNotFoundError(f"missing {len(missing_files)} checkpoint shards, first: {sorted(missing_files)[0]}")
-  model = Transformer(kimi_k3_config(max_context))
-  _shard_kimi_k3(model, tuple(f"{Device.DEFAULT}:{i}" for i in range(devices)))
-  consumed = _load_nonexperts(root, weight_map, model, progress)
-  # Expert staging graphs are acyclic and released by reference counting after each layer. Avoid
-  # unnecessary cyclic-collector scans across the complete persistent model graph during this loop.
-  gc_was_enabled = gc.isenabled()
-  gc.disable()
-  try: consumed.update(_load_experts(root, weight_map, model, progress))
-  finally:
-    if gc_was_enabled: gc.enable()
-  unused_language = {k for k in weight_map if k.startswith("language_model.")} - consumed
-  if unused_language: raise ValueError(f"unmapped language tensors: {sorted(unused_language)[:20]}")
-  return model
-
-__all__ = ["KIMI_K3_FULL_ATTN_LAYERS", "KIMI_K3_SSM_LAYERS", "KIMI_K3_TEXT_SIZE", "KIMI_K3_TP8_BYTES_PER_GPU",
-           "audit_kimi_k3_checkpoint", "kimi_k3_config", "kimi_k3_smoke_config", "load_kimi_k3", "load_kimi_tokenizer_data"]

tinygrad/llm/model.py

@@ -1,16 +1,9 @@
 from __future__ import annotations
-import array, functools, itertools, pathlib
+import functools, itertools, pathlib
 from dataclasses import dataclass, replace
-from typing import Callable, cast
-from tinygrad import Tensor, nn, UOp, TinyJit, getenv, function, dtypes
-from tinygrad.device import MultiBuffer
+from tinygrad import Tensor, nn, UOp, TinyJit, getenv, function
 from tinygrad.nn import Linear
 from tinygrad.llm.gguf import gguf_load
-from tinygrad.llm.quant import dequantize_mxfp4, quantize_dequantize_mxfp8
-from tinygrad.llm.kernels import amd_custom_kernels_supported, amd_exact_bf16_custom_kernels_supported, amd_int32_item, \
-  amd_packed_mxfp4_supported, amd_wave64_custom_kernels_supported, bf16_matvec, bf16_mfma_splitk, bf16_partial_linear, \
-  dual_bf16_matvec, dual_input_bf16_matvec, \
-  gated_delta_prefill, kda_fgb_linear, kda_qkv_linear, mxfp4_expert_linear, mxfp8_quantize_dequantize
 from tinygrad.uop.ops import resolve
 
 @functools.cache
@@ -27,37 +20,12 @@ class ExpertWeights:
     # sel: (B, T, k), x: (B, T, 1, in) or (B, T, k, in) -> output: (B, T, k, out)
     return (x.unsqueeze(-2) @ self.weight[sel].transpose(-1, -2)).contiguous().squeeze(-2)
 
-class MXFP4ExpertWeights:
-  """Routed-expert weights stored as packed OCP MXFP4 with one E8M0 scale per 32 values."""
-  def __init__(self, num_experts:int, in_features:int, out_features:int):
-    if in_features % 32: raise ValueError(f"MXFP4 expert input size must be divisible by 32, got {in_features}")
-    self.in_features, self.out_features = in_features, out_features
-    self.weight = Tensor.zeros(num_experts, out_features, in_features//2, dtype=dtypes.uint8)
-    self.weight_scale = Tensor.full((num_experts, out_features, in_features//32), 127, dtype=dtypes.uint8)
-  def __call__(self, sel:Tensor, x:Tensor, quantized:bool=False, partial:bool=False) -> Tensor:
-    # Only selected weights are expanded, so packed storage remains resident during generation.
-    if isinstance(self.weight.device, tuple) and not isinstance(sel.device, tuple): sel = sel.shard(self.weight.device, axis=None)
-    if not quantized:
-      x = mxfp8_quantize_dequantize(x.cast(dtypes.bfloat16)) if amd_custom_kernels_supported(x.device) else \
-        quantize_dequantize_mxfp8(x.cast(dtypes.bfloat16))
-    # gfx11 has no native FP4 instructions, but decoding nibbles inside the dot product still avoids
-    # the much larger selected-expert BF16 temporary. Gate/up weights are output-sharded in TP.
-    if amd_packed_mxfp4_supported(self.weight.device):
-      return mxfp4_expert_linear(sel, x, self.weight, self.weight_scale, partial=partial)
-    weight = dequantize_mxfp4(self.weight[sel], self.weight_scale[sel], dtype=dtypes.bfloat16)
-    return (x.unsqueeze(-2) @ weight.transpose(-1, -2)).contiguous().squeeze(-2)
-
 def apply_rope(x:Tensor, freqs_cis:Tensor) -> Tensor:
   assert x.shape[-1] % 2 == 0
   cos, sin = freqs_cis.reshape(1, 1, x.shape[2], -1).chunk(2, dim=-1)
   x1, x2 = x.chunk(2, dim=-1)
   return (x1 * cos - x2 * sin).cat(x2 * cos + x1 * sin, dim=-1)
 
-def l2norm(x:Tensor, eps:float=1e-6) -> Tensor:
-  """FLA-compatible L2 normalization: FP32 reduction and epsilon inside the square root."""
-  dtype, x = x.dtype, x.float()
-  return (x * (x.square().sum(axis=-1, keepdim=True, dtype=dtypes.float32) + eps).rsqrt()).cast(dtype)
-
 def pairwise_topk(x: Tensor, k: int) -> tuple[Tensor, Tensor]:
   n = x.shape[-1]
   vals = Tensor.arange(n).reshape(1,1,n).cast(x.dtype).expand(x.shape)
@@ -66,16 +34,6 @@ def pairwise_topk(x: Tensor, k: int) -> tuple[Tensor, Tensor]:
   sel = x.const_like(0).scatter(-1, cmp.sum(axis=-1).cast('int32'), vals)[:,:,n-k:].cast('int32')
   return x.gather(-1, sel), sel
 
-def iterative_topk(x:Tensor, k:int) -> tuple[Tensor, Tensor]:
-  """O(k*N) top-k for very wide MoE routers, with stable first-index tie breaking."""
-  work, values, indices = x, [], []
-  for _ in range(k):
-    sel = work.argmax(-1, keepdim=True)
-    values.append(x.gather(-1, sel))
-    indices.append(sel)
-    work = work.scatter(-1, sel, x.dtype.min)
-  return values[0].cat(*values[1:], dim=-1), indices[0].cat(*indices[1:], dim=-1)
-
 @dataclass(frozen=True)
 class SSMConfig:
   conv_kernel: int
@@ -84,7 +42,6 @@ class SSMConfig:
   time_step_rank: int
   inner_size: int
   kda: bool = False
-  channel_decay: bool = False
 
 @dataclass(frozen=True)
 class TransformerConfig:
@@ -116,20 +73,6 @@ class TransformerConfig:
   routed_scaling_factor: float = 1.0
   qkv_bias: bool = False
   expert_bias: bool = False
-  expert_mxfp4: bool = False
-  bf16_activations: bool = False
-  kda_split_qkv: bool = False
-  # Kimi K3 extensions. Defaults preserve all existing model behavior.
-  activation_situ_beta: float = 0.0
-  activation_situ_linear_beta: float = 0.0
-  routed_expert_dim: int = 0
-  latent_moe_norm: bool = False
-  route_weights_uncorrected: bool = False
-  attn_res_block_size: int = 0
-  kda_full_rank_gate: bool = False
-  kda_gate_lower_bound: float = 0.0
-  recurrent_prefill_chunked: bool = False
-  recurrent_prefill_chunk_size: int = 0
 
 class FFNBlock:
   def __init__(self, config:TransformerConfig):
@@ -143,15 +86,9 @@ class FFNBlock:
     if config.num_experts > 0:
       self.ffn_gate_inp = Linear(config.dim, config.num_experts, bias=False)  # router
       if config.expert_bias: self.exp_probs_b = {"bias": Tensor.zeros(config.num_experts)}
-      expert_cls = MXFP4ExpertWeights if config.expert_mxfp4 else ExpertWeights
-      expert_dim = config.routed_expert_dim or config.dim
-      self.ffn_gate_exps = expert_cls(config.num_experts, expert_dim, config.hidden_dim)
-      self.ffn_up_exps = expert_cls(config.num_experts, expert_dim, config.hidden_dim)
-      self.ffn_down_exps = expert_cls(config.num_experts, config.hidden_dim, expert_dim)
-      if config.routed_expert_dim:
-        self.ffn_routed_down = Linear(config.dim, expert_dim, bias=False)
-        self.ffn_routed_up = Linear(expert_dim, config.dim, bias=False)
-        if config.latent_moe_norm: self.ffn_routed_norm = nn.RMSNorm(expert_dim, config.norm_eps)
+      self.ffn_gate_exps = ExpertWeights(config.num_experts, config.dim, config.hidden_dim)
+      self.ffn_up_exps = ExpertWeights(config.num_experts, config.dim, config.hidden_dim)
+      self.ffn_down_exps = ExpertWeights(config.num_experts, config.hidden_dim, config.dim)
       if config.shared_expert_dim > 0:
         self.ffn_gate_shexp = Linear(config.dim, config.shared_expert_dim, bias=False)
         self.ffn_up_shexp = Linear(config.dim, config.shared_expert_dim, bias=False)
@@ -162,85 +99,28 @@ class FFNBlock:
       self.ffn_up      = Linear(config.dim, config.hidden_dim, bias=False)
       self.ffn_down    = Linear(config.hidden_dim, config.dim, bias=False)
 
-    if config.attn_res_block_size:
-      self.attn_res_norm, self.mlp_res_norm = nn.RMSNorm(config.dim, config.norm_eps), nn.RMSNorm(config.dim, config.norm_eps)
-      self.attn_res_proj, self.mlp_res_proj = Linear(config.dim, 1, bias=False), Linear(config.dim, 1, bias=False)
-
-  def _activation(self, gate:Tensor, up:Tensor) -> Tensor:
-    if not self.config.activation_situ_beta: return gate.silu() * up
-    gate32, up32, beta = gate.float(), up.float(), self.config.activation_situ_beta
-    gate32 = beta * (gate32 / beta).tanh() * gate32.sigmoid()
-    if (linear_beta := self.config.activation_situ_linear_beta): up32 = linear_beta * (up32 / linear_beta).tanh()
-    return (gate32 * up32).cast(gate.dtype)
-
   def _feed_forward(self, x:Tensor) -> Tensor:
     if hasattr(self, 'ffn_gate_exps'):
       h = x.unsqueeze(2)  # (B, T, 1, D) - add expert dim for broadcasting
-      # Kimi computes router logits in FP32 even though the residual stream and weights are BF16.
-      logits = x.float().linear(self.ffn_gate_inp.weight.float().transpose()) if self.config.bf16_activations else self.ffn_gate_inp(x)
+      logits = self.ffn_gate_inp(x)
       if hasattr(self, 'exp_probs_b'):
-        scores = logits.sigmoid()
-        adjusted_scores = scores + self.exp_probs_b["bias"]
-        topk = iterative_topk if self.config.num_experts >= 512 else pairwise_topk
-        _, sel = topk(adjusted_scores, self.config.num_experts_per_tok)
-        probs = (scores if self.config.route_weights_uncorrected else adjusted_scores).gather(-1, sel)
-        # Kimi-Linear-48B's older reference weights corrected scores. K3 selects with the correction
-        # but gathers the uncorrected sigmoid scores, so keep this an explicit compatibility switch.
-        if self.config.norm_topk_prob: probs = probs / (probs.sum(axis=-1, keepdim=True) + 1e-20)
+        probs = logits.sigmoid()
+        _, sel = pairwise_topk(probs + self.exp_probs_b["bias"], self.config.num_experts_per_tok)
+        probs = probs.gather(-1, sel)
+        if self.config.norm_topk_prob: probs = probs / probs.sum(axis=-1, keepdim=True)
       else:
         vals, sel = pairwise_topk(logits, self.config.num_experts_per_tok)
         probs = vals.softmax(-1) if self.config.norm_topk_prob else logits.softmax(-1).gather(-1, sel)
       probs = probs * self.config.routed_scaling_factor
-      if hasattr(self, 'ffn_routed_down'): h = self.ffn_routed_down(x).unsqueeze(2)
-      if isinstance(self.ffn_gate_exps, MXFP4ExpertWeights) and amd_packed_mxfp4_supported(h.device):
-        hq = mxfp8_quantize_dequantize(h.cast(dtypes.bfloat16)) if amd_custom_kernels_supported(h.device) else \
-          quantize_dequantize_mxfp8(h.cast(dtypes.bfloat16))
-        gate = self.ffn_gate_exps(sel, hq, quantized=True)
-        up = cast(MXFP4ExpertWeights, self.ffn_up_exps)(sel, hq, quantized=True)
-      else: gate, up = self.ffn_gate_exps(sel, h), self.ffn_up_exps(sel, h)
-      routed_activation = self._activation(gate, up).contiguous()
-      combine_down = resolve(x.shape[1] == 1) and isinstance(self.ffn_down_exps, MXFP4ExpertWeights) and \
-        hasattr(self, 'ffn_gate_shexp') and not hasattr(self, 'ffn_routed_up') and amd_custom_kernels_supported(x.device)
-      x_down = cast(MXFP4ExpertWeights, self.ffn_down_exps)(sel, routed_activation, partial=True) if combine_down else \
-        self.ffn_down_exps(sel, routed_activation)
-      out = (x_down * probs.unsqueeze(-1).unsqueeze(-1)).sum(axis=2) if combine_down else \
-        (x_down * probs.unsqueeze(-1)).sum(axis=2).cast(x_down.dtype)  # (B, T, D[, devices])
-      combine_final = resolve(x.shape[1] == 1) and hasattr(self, 'ffn_routed_up') and hasattr(self, 'ffn_gate_shexp') and \
-        not hasattr(self, 'ffn_gate_inp_shexp') and isinstance(self.ffn_routed_up.weight.device, tuple) and \
-        isinstance(self.ffn_down_shexp.weight.device, tuple) and self.ffn_routed_up.weight.uop.axis == self.ffn_down_shexp.weight.uop.axis == 1 and \
-        self.ffn_routed_up.weight.shape[1] % (32*len(self.ffn_routed_up.weight.device)) == 0 and \
-        self.ffn_down_shexp.weight.shape[1] % (32*len(self.ffn_down_shexp.weight.device)) == 0 and \
-        amd_exact_bf16_custom_kernels_supported(x.device)
-      if hasattr(self, 'ffn_routed_up'):
-        if hasattr(self, 'ffn_routed_norm'): out = self.ffn_routed_norm(out)
-        out = bf16_partial_linear(out, self.ffn_routed_up.weight) if combine_final else self.ffn_routed_up(out)
+      x_down = self.ffn_down_exps(sel, (self.ffn_gate_exps(sel, h).silu() * self.ffn_up_exps(sel, h)).contiguous())  # (B, T, k, D)
+      out = (x_down * probs.unsqueeze(-1)).sum(axis=2)  # (B, T, D)
       if hasattr(self, 'ffn_gate_shexp'):
-        if resolve(x.shape[1] == 1) and amd_exact_bf16_custom_kernels_supported(x.device) and \
-           self.ffn_gate_shexp.weight.shape == self.ffn_up_shexp.weight.shape:
-          shared_gate, shared_up = dual_bf16_matvec(x, self.ffn_gate_shexp.weight, self.ffn_up_shexp.weight,
-                                                    fast=amd_custom_kernels_supported(x.device))
-        else: shared_gate, shared_up = self.ffn_gate_shexp(x).contiguous(), self.ffn_up_shexp(x).contiguous()
-        shared_activation = self._activation(shared_gate, shared_up).contiguous()
-        if combine_down:
-          out = (out + bf16_partial_linear(shared_activation, self.ffn_down_shexp.weight)).sum(3).cast(dtypes.bfloat16)
-        elif combine_final:
-          out = (out + bf16_partial_linear(shared_activation, self.ffn_down_shexp.weight)).sum(3).cast(dtypes.bfloat16)
-        else:
-          shexp = self.ffn_down_shexp(shared_activation)
-          if hasattr(self, 'ffn_gate_inp_shexp'):
-            shexp = shexp * (x * self.ffn_gate_inp_shexp["weight"]).sum(axis=-1, keepdim=True).sigmoid()
-          out = out + shexp
+        shexp = self.ffn_down_shexp(self.ffn_gate_shexp(x).silu().contiguous() * self.ffn_up_shexp(x))
+        if hasattr(self, 'ffn_gate_inp_shexp'): shexp = shexp * (x * self.ffn_gate_inp_shexp["weight"]).sum(axis=-1, keepdim=True).sigmoid()
+        out = out + shexp
       return out
     # TODO: remove the need for this contiguous
-    if resolve(x.shape[1] == 1) and amd_exact_bf16_custom_kernels_supported(x.device) and \
-       self.ffn_gate.weight.shape == self.ffn_up.weight.shape:
-      dense_gate, dense_up = dual_bf16_matvec(x, self.ffn_gate.weight, self.ffn_up.weight, fast=amd_custom_kernels_supported(x.device))
-    else: dense_gate, dense_up = self.ffn_gate(x).contiguous(), self.ffn_up(x).contiguous()
-    dense_activation = self._activation(dense_gate, dense_up).contiguous()
-    if resolve(x.shape[1] == 1) and isinstance(self.ffn_down.weight.device, tuple) and self.ffn_down.weight.uop.axis == 1 and \
-       self.ffn_down.weight.shape[1] % (32*len(self.ffn_down.weight.device)) == 0 and amd_exact_bf16_custom_kernels_supported(x.device):
-      return bf16_partial_linear(dense_activation, self.ffn_down.weight).sum(3).cast(dtypes.bfloat16)
-    return self.ffn_down(dense_activation)
+    return self.ffn_down(self.ffn_gate(x).silu().contiguous() * self.ffn_up(x))
 
   # given the token-prefix match, return how much cached state this block can still reuse
   def _reusable_prefix_len(self, prefix_len:int, cached_len:int) -> int: return prefix_len
@@ -251,11 +131,6 @@ class FFNBlock:
 
   def __call__(self, x: Tensor, start_pos: int|UOp):
     self._init_state(x)
-    # Kimi's heterogeneous TP shards are captured by the outer TinyJit; per-block precompilation
-    # cannot represent their differently shaped local buffers as one implicit parameter bundle.
-    if self.config.bf16_activations:
-      h = x + self._attention(self.attn_norm(x), start_pos)
-      return (h + self._feed_forward(self.ffn_norm(h))).contiguous()
     # we pass in the weights implicitly so we unpack the GGUF on the fly
     @function(precompile=True, allow_implicit=True)
     def _run(x:Tensor, start_pos:int|UOp):
@@ -263,32 +138,6 @@ class FFNBlock:
       return (h + self._feed_forward(self.ffn_norm(h))).contiguous()
     return _run(x, start_pos)
 
-  @staticmethod
-  def _apply_attn_res(prefix_sum:Tensor, block_residual:Tensor, proj:Linear, norm:nn.RMSNorm) -> Tensor:
-    # Both inputs are flattened over B*T. Scoring is intentionally FP32, matching K3 eager inference.
-    v = block_residual.cat(prefix_sum.unsqueeze(1), dim=1)
-    vf = v.float()
-    k = vf * (vf.square().mean(axis=-1, keepdim=True) + norm.eps).rsqrt()
-    assert norm.weight is not None
-    scores = (k * (norm.weight.float() * proj.weight.squeeze(0).float())).sum(axis=-1)
-    return (scores.softmax(-1).unsqueeze(1) @ vf).squeeze(1).cast(v.dtype)
-
-  def attn_residual(self, x:Tensor, start_pos:int|UOp, block_residual:Tensor, layer_idx:int) -> tuple[Tensor, Tensor]:
-    self._init_state(x)
-    shape, prefix_sum = x.shape, x
-    prefix:Tensor|None = prefix_sum
-    if block_residual.shape[1]: x = self._apply_attn_res(x.reshape(-1, shape[-1]), block_residual,
-                                                        self.attn_res_proj, self.attn_res_norm).reshape(shape)
-    if layer_idx % self.config.attn_res_block_size == 0:
-      block_residual = block_residual.cat(prefix_sum.reshape(-1, shape[-1]).unsqueeze(1), dim=1)
-      prefix = None
-    attn = self._attention(self.attn_norm(x), start_pos)
-    prefix = attn if prefix is None else prefix + attn
-    x = self._apply_attn_res(prefix.reshape(-1, shape[-1]), block_residual,
-                             self.mlp_res_proj, self.mlp_res_norm).reshape(shape)
-    mlp = self._feed_forward(self.ffn_norm(x))
-    return (prefix + mlp).contiguous(), block_residual
-
 class TransformerBlock(FFNBlock):
   def __init__(self, config:TransformerConfig):
     super().__init__(config)
@@ -330,9 +179,7 @@ class TransformerBlock(FFNBlock):
 
     # NOTE: this mask is causal_lower_right, not the causal_upper_left generated by is_casual = True
     # TODO: this if statement should be removed and it shouldn't generate extra kernels
-    # Build the static T×T causal corner on-device, then prepend the unmasked cached prefix.
-    # A broadcast const with symbolic width otherwise defaults to CPU in multi-device graphs.
-    mask = Tensor.full((1, 1, T, T), float("-inf"), dtype=x.dtype, device=x.device).triu(1).pad(((0, 0),)*3+((start_pos, 0),)) \
+    mask = Tensor.full((1, 1, T, start_pos+T), float("-inf"), dtype=x.dtype, buffer=False).triu(start_pos+1) \
       if resolve(T != 1) else None
     attn = q.scaled_dot_product_attention(k, v, attn_mask=mask, enable_gqa=True)     # (B,H,T,Hd)
     attn = attn.transpose(1, 2).reshape(B, T, -1)                                    # back to (B,T,D)
@@ -341,8 +188,7 @@ class TransformerBlock(FFNBlock):
   def _init_state(self, x:Tensor):
     if not hasattr(self, "cache_kv"):
       # TODO: how is the dtype of this determined?
-      self.cache_kv = Tensor.empty(2, x.shape[0], self.config.n_kv_heads, self.config.max_context, self.config.head_dim,
-                                   device=x.device, dtype=x.dtype)
+      self.cache_kv = Tensor.empty(2, x.shape[0], self.config.n_kv_heads, self.config.max_context, self.config.head_dim, device=x.device)
       self.freqs_cis = precompute_freqs_cis(self.config.rope_dim, self.config.max_context, self.config.rope_theta, device=x.device)
 
 class MLATransformerBlock(FFNBlock):
@@ -360,23 +206,17 @@ class MLATransformerBlock(FFNBlock):
     self.attn_k_b = {"weight": Tensor.zeros(config.n_heads, config.kv_lora_rank, qk_nope_head_dim)}
     self.attn_v_b = {"weight": Tensor.zeros(config.n_heads, config.v_head_dim, config.kv_lora_rank)}
     self.attn_output = Linear(config.n_heads * config.v_head_dim, config.dim, bias=False)
-    if config.attn_output_gate: self.attn_gate = Linear(config.dim, config.n_heads * config.v_head_dim, bias=False)
 
   def _attention(self, x:Tensor, start_pos:int|UOp) -> Tensor:
     B, T, _ = x.shape
     q_nope_head_dim = self.config.head_dim - self.config.rope_dim
-    mfma_decode = resolve(T == 1) and x.shape[-1] % 256 == 0 and amd_wave64_custom_kernels_supported(x.device)
-    q_a_mfma = self.config.q_lora_rank > 0 and mfma_decode and self.attn_q_a.weight.shape[0] % 16 == 0
-    q_a = bf16_mfma_splitk(x, self.attn_q_a.weight) if q_a_mfma else \
-      self.attn_q_a(x) if self.config.q_lora_rank > 0 else None
-    q_proj = self.attn_q_b(self.attn_q_a_norm(q_a)) if q_a is not None else self.attn_q(x)
+    q_proj = self.attn_q_b(self.attn_q_a_norm(self.attn_q_a(x))) if self.config.q_lora_rank > 0 else self.attn_q(x)
     q = q_proj.reshape(B, T, self.config.n_heads, self.config.head_dim).transpose(1, 2)
     q_nope, q_rope = q[..., :q_nope_head_dim], q[..., q_nope_head_dim:]
     if not self.config.ssm or not self.config.ssm.kda: q_rope = apply_rope(q_rope, self.freqs_cis[start_pos:start_pos+T])
     q = (q_nope @ self.attn_k_b["weight"].transpose(-1, -2)).cat(q_rope, dim=-1)
 
-    kv_a_mfma = mfma_decode and self.attn_kv_a_mqa.weight.shape[0] % 16 == 0
-    kv_a = bf16_mfma_splitk(x, self.attn_kv_a_mqa.weight) if kv_a_mfma else self.attn_kv_a_mqa(x)
+    kv_a = self.attn_kv_a_mqa(x)
     c_kv = self.attn_kv_a_norm(kv_a[..., :self.config.kv_lora_rank])
     k_rope = kv_a[..., self.config.kv_lora_rank:].reshape(B, T, 1, self.config.rope_dim).transpose(1, 2)
     if not self.config.ssm or not self.config.ssm.kda: k_rope = apply_rope(k_rope, self.freqs_cis[start_pos:start_pos+T])
@@ -385,23 +225,17 @@ class MLATransformerBlock(FFNBlock):
     k = Tensor(self.cache_k.uop.after(self.cache_k[:, :, start_pos:start_pos+T, :].uop.store(k_store.uop)))[:, :, 0:start_pos+T, :]
     v = k[..., :self.config.kv_lora_rank]
 
-    mask = Tensor.full((1, 1, T, T), float("-inf"), dtype=x.dtype, device=x.device).triu(1).pad(((0, 0),)*3+((start_pos, 0),)) \
+    mask = Tensor.full((1, 1, T, start_pos+T), float("-inf"), dtype=x.dtype, buffer=False).triu(start_pos+1) \
       if resolve(T != 1) else None
     attn = q @ k.transpose(-1, -2) * (1.0 / self.config.head_dim ** 0.5)
     if mask is not None: attn = attn + mask
-    # Match eager Kimi MLA: normalize attention scores in FP32, then return to the query dtype.
-    attn = attn.softmax(-1, dtype=dtypes.float32).cast(q.dtype)
+    attn = attn.softmax(-1)
     attn = ((attn @ v) @ self.attn_v_b["weight"].transpose(-1, -2)).transpose(1, 2).reshape(B, T, -1)
-    if hasattr(self, "attn_gate"): attn = attn * self.attn_gate(x).sigmoid()
-    if resolve(T == 1) and isinstance(self.attn_output.weight.device, tuple) and \
-       self.attn_output.weight.shape[1] % (32*len(self.attn_output.weight.device)) == 0 and amd_exact_bf16_custom_kernels_supported(attn.device):
-      return bf16_partial_linear(attn, self.attn_output.weight).sum(3).cast(dtypes.bfloat16)
     return self.attn_output(attn)
 
   def _init_state(self, x:Tensor):
     if not hasattr(self, "cache_k"):
-      self.cache_k = Tensor.empty(x.shape[0], 1, self.config.max_context, self.config.kv_lora_rank + self.config.rope_dim,
-                                  device=x.device, dtype=x.dtype)
+      self.cache_k = Tensor.empty(x.shape[0], 1, self.config.max_context, self.config.kv_lora_rank + self.config.rope_dim, device=x.device)
       self.freqs_cis = precompute_freqs_cis(self.config.rope_dim, self.config.max_context, self.config.rope_theta, device=x.device)
 
 class GatedDeltaNetBlock(FFNBlock):
@@ -411,144 +245,68 @@ class GatedDeltaNetBlock(FFNBlock):
     assert self.num_v_heads % self.num_k_heads == 0
     self.head_v_dim, self.ssm_conv_kernel = ssm.inner_size // ssm.time_step_rank, ssm.conv_kernel
     self.conv_channels, self.q_dim = ssm.inner_size + 2*ssm.group_count*ssm.state_size, ssm.state_size*ssm.group_count
-    if ssm.kda and config.kda_split_qkv:
-      self.attn_q, self.attn_k = Linear(config.dim, self.q_dim, bias=False), Linear(config.dim, self.q_dim, bias=False)
-      self.attn_v = Linear(config.dim, ssm.inner_size, bias=False)
-      self.ssm_q_conv1d = {"weight": Tensor.zeros(self.q_dim, self.ssm_conv_kernel)}
-      self.ssm_k_conv1d = {"weight": Tensor.zeros(self.q_dim, self.ssm_conv_kernel)}
-      self.ssm_v_conv1d = {"weight": Tensor.zeros(ssm.inner_size, self.ssm_conv_kernel)}
-    else:
-      self.attn_qkv = Linear(config.dim, self.conv_channels, bias=False)
-      self.ssm_conv1d = {"weight": Tensor.zeros(self.conv_channels, self.ssm_conv_kernel)}
+    self.attn_qkv = Linear(config.dim, self.conv_channels, bias=False)
     if ssm.kda:
-      if config.kda_full_rank_gate: self.ssm_g_full = Linear(config.dim, ssm.inner_size, bias=False)
-      else: self.ssm_g_a, self.ssm_g_b = Linear(config.dim, self.head_v_dim, bias=False), Linear(self.head_v_dim, ssm.inner_size, bias=False)
+      self.ssm_g_a, self.ssm_g_b = Linear(config.dim, self.head_v_dim, bias=False), Linear(self.head_v_dim, ssm.inner_size, bias=False)
       self.ssm_f_a, self.ssm_f_b = Linear(config.dim, self.head_k_dim, bias=False), Linear(self.head_k_dim, ssm.inner_size, bias=False)
     else:
       self.attn_gate = Linear(config.dim, ssm.inner_size, bias=False)
       self.ssm_alpha = Linear(config.dim, self.num_v_heads, bias=False)
     self.ssm_beta = Linear(config.dim, self.num_v_heads, bias=False)
+    self.ssm_conv1d = {"weight": Tensor.zeros(self.conv_channels, self.ssm_conv_kernel)}
     self.ssm_dt = {"bias": Tensor.zeros(ssm.inner_size if ssm.kda else self.num_v_heads)}
-    self.ssm_a = Tensor.zeros(self.head_v_dim if ssm.channel_decay else self.num_v_heads, 1) if ssm.kda else Tensor.zeros(self.num_v_heads)
-    self.kda_channel_decay = ssm.channel_decay
+    self.ssm_a = Tensor.zeros(self.num_v_heads, 1) if ssm.kda else Tensor.zeros(self.num_v_heads)
     self.ssm_norm, self.ssm_out = nn.RMSNorm(self.head_v_dim, config.norm_eps), Linear(ssm.inner_size, config.dim, bias=False)
 
   def _attention(self, x:Tensor, start_pos:int|UOp) -> Tensor:
     B, T, _ = x.shape
+    assert T == 1, "GatedDeltaNetBlock currently only supports T=1"
 
     # input processing
-    # Kimi-Linear is a BF16 model. Qwen 3.5 GGDN checkpoints historically use FP16 here.
-    x = x.cast(dtypes.bfloat16) if self.config.ssm and self.config.ssm.kda else x.half()
-    fused_fg = hasattr(self, "ssm_g_a") and resolve(T == 1) and amd_custom_kernels_supported(x.device) and \
-      self.ssm_g_a.weight.shape == self.ssm_f_a.weight.shape
-    mfma_decode = resolve(T == 1) and x.shape[-1] % 256 == 0 and amd_wave64_custom_kernels_supported(x.device)
-    if hasattr(self, "ssm_g_full"): out_gate = self.ssm_g_full(x)
-    elif hasattr(self, "ssm_g_a"):
-      if fused_fg:
-        gate_a, alpha_a, beta_logits = kda_fgb_linear(x, self.ssm_g_a.weight, self.ssm_f_a.weight, self.ssm_beta.weight)
-        out_gate, alpha_logits = dual_input_bf16_matvec(gate_a, alpha_a, self.ssm_g_b.weight, self.ssm_f_b.weight)
-      else: out_gate = self.ssm_g_b(self.ssm_g_a(x))
-    else: out_gate = self.attn_gate(x)
-    if not fused_fg: beta_logits = self.ssm_beta(x)
-    if not fused_fg:
-      if hasattr(self, "ssm_f_a"):
-        f_a = bf16_mfma_splitk(x, self.ssm_f_a.weight) if mfma_decode and self.ssm_f_a.weight.shape[0] % 16 == 0 else self.ssm_f_a(x)
-        alpha_logits = self.ssm_f_b(f_a)
-      else: alpha_logits = self.ssm_alpha(x)
+    x = x.half()
+    out_gate = self.ssm_g_b(self.ssm_g_a(x)) if hasattr(self, "ssm_g_a") else self.attn_gate(x)
+    out_gate = out_gate.reshape(B, 1, self.num_v_heads, self.head_v_dim)
+    beta = self.ssm_beta(x).sigmoid().reshape(B, self.num_v_heads, 1, 1)
+    alpha = self.ssm_f_b(self.ssm_f_a(x)) if hasattr(self, "ssm_f_a") else self.ssm_alpha(x)
+    alpha = ((alpha.float() + self.ssm_dt["bias"]).softplus().reshape(B, self.num_v_heads, -1) *
+             self.ssm_a.reshape(1, self.num_v_heads, -1)).exp().unsqueeze(-2)
 
-    # Causal depthwise Q/K/V convolution. All tokens are projected together, then the recurrent
-    # update is fused into one kernel so prefill doesn't build a Python-unrolled graph.
-    split_qkv = hasattr(self, "attn_q")
-    if split_qkv:
-      if resolve(T == 1) and amd_packed_mxfp4_supported(x.device) and \
-         self.attn_q.weight.shape == self.attn_k.weight.shape == self.attn_v.weight.shape:
-        projected_q, projected_k, projected_v = kda_qkv_linear(x, self.attn_q.weight, self.attn_k.weight, self.attn_v.weight)
-      else: projected_q, projected_k, projected_v = self.attn_q(x), self.attn_k(x), self.attn_v(x)
-      # Snapshot mutable caches before constructing the recurrence. Otherwise the final store can
-      # overwrite their buffers before earlier outputs in a multi-token lazy graph consume them.
-      conv_state_q, conv_state_k, conv_state_v = self.conv_state_q.clone(), self.conv_state_k.clone(), self.conv_state_v.clone()
-    else: projected, conv_state = self.attn_qkv(x), self.conv_state
-    def causal_conv(projected:Tensor, state:Tensor, weight:Tensor) -> tuple[Tensor, Tensor]:
-      window = state.cat(projected, dim=1)
-      out = functools.reduce(lambda a,b: a+b, (window[:, i:i+T] * weight[:, i] for i in range(self.ssm_conv_kernel))).silu()
-      return out, window[:, T:T+self.ssm_conv_kernel-1]
-    if split_qkv:
-      q, conv_state_q = causal_conv(projected_q, conv_state_q, self.ssm_q_conv1d["weight"])
-      k, conv_state_k = causal_conv(projected_k, conv_state_k, self.ssm_k_conv1d["weight"])
-      v, conv_state_v = causal_conv(projected_v, conv_state_v, self.ssm_v_conv1d["weight"])
-    else:
-      conv_out, conv_state = causal_conv(projected, conv_state, self.ssm_conv1d["weight"])
-      q, k, v = conv_out.split([self.q_dim, self.q_dim, self.conv_channels - 2*self.q_dim], dim=-1)
+    # qkv conv
+    conv_window = self.conv_state.cat(self.attn_qkv(x), dim=1)
+    conv_out = (conv_window * self.ssm_conv1d["weight"].T.unsqueeze(0)).sum(1).silu()
+    q, k, v = conv_out.split([self.q_dim, self.q_dim, self.conv_channels - 2*self.q_dim], dim=-1)
+    q = q.reshape(B, self.num_k_heads, self.head_k_dim).normalize(dim=-1).repeat(1, self.num_v_heads//self.num_k_heads, 1)
+    k = k.reshape(B, self.num_k_heads, self.head_k_dim).normalize(dim=-1).repeat(1, self.num_v_heads//self.num_k_heads, 1)
+    v = v.reshape(B, self.num_v_heads, self.head_v_dim)
+    q, k, v = q.mul(self.head_k_dim**-0.5).unsqueeze(-1), k.unsqueeze(-1), v.unsqueeze(-1)
 
-    q, k = q.reshape(B, T, self.num_k_heads, self.head_k_dim), k.reshape(B, T, self.num_k_heads, self.head_k_dim)
-    q, k = (l2norm(q), l2norm(k)) if self.config.ssm and self.config.ssm.kda else (q.normalize(dim=-1), k.normalize(dim=-1))
-    q = q.repeat(1, 1, self.num_v_heads//self.num_k_heads, 1).transpose(1, 2).float() * self.head_k_dim**-0.5
-    k = k.repeat(1, 1, self.num_v_heads//self.num_k_heads, 1).transpose(1, 2).float()
-    v = v.reshape(B, T, self.num_v_heads, self.head_v_dim).transpose(1, 2).float()
-    beta = (beta_logits.float() if self.config.ssm and self.config.ssm.kda else beta_logits).sigmoid().transpose(1, 2)
-    gate_logits = (alpha_logits.float() + self.ssm_dt["bias"]).reshape(B, T, self.num_v_heads, -1)
-    a_shape = (1, 1, 1, self.head_v_dim) if self.kda_channel_decay else (1, 1, self.num_v_heads, 1)
-    if self.config.kda_gate_lower_bound:
-      log_alpha = self.config.kda_gate_lower_bound * ((-self.ssm_a).reshape(a_shape) * gate_logits).sigmoid()
-    else: log_alpha = gate_logits.softplus() * self.ssm_a.reshape(a_shape)
-    alpha = log_alpha.squeeze(-1).transpose(1, 2).exp() if log_alpha.shape[-1] == 1 else log_alpha.permute(0, 2, 1, 3).exp()
-    if T == 1:
-      # Keep decode on the small elementwise graph. The fused prefill kernel writes a temporary
-      # recurrent matrix, which is worthwhile for multiple tokens but needlessly copies state at T=1.
-      decay = alpha if len(alpha.shape) == 4 else alpha.unsqueeze(-1)
-      recurrent_state = self.recurrent_state * decay
-      k1, q1 = k[:, :, 0].unsqueeze(-1), q[:, :, 0].unsqueeze(-1)
-      recurrent_state = recurrent_state + ((v[:, :, 0].unsqueeze(-1) - recurrent_state@k1) * beta[:, :, 0].reshape(B, self.num_v_heads, 1, 1)) @ \
-                        k1.transpose(-1, -2)
-      core = (recurrent_state @ q1).squeeze(-1).unsqueeze(2)
-    else: core, recurrent_state = gated_delta_prefill(q, k, v, beta, alpha, self.recurrent_state)
-    core = core.transpose(1, 2)
+    # recurrent
+    recurrent_state = self.recurrent_state * alpha
+    recurrent_state = recurrent_state + ((v - recurrent_state@k) * beta)@k.transpose(-1, -2)
 
-    # Store each cache with its own AFTER. Multi-device lowering handles one sharded STORE per
-    # AFTER; grouping these effects under one cache silently drops stores on the other shards.
-    state_updates:list[Tensor]
-    if split_qkv:
-      state_updates = [self.conv_state_q.assign(conv_state_q.cast(self.conv_state_q.dtype)),
-                       self.conv_state_k.assign(conv_state_k.cast(self.conv_state_k.dtype)),
-                       self.conv_state_v.assign(conv_state_v.cast(self.conv_state_v.dtype))]
-    else: state_updates = [self.conv_state.assign(conv_state.cast(self.conv_state.dtype))]
-    state_updates.append(self.recurrent_state.assign(recurrent_state.cast(self.recurrent_state.dtype)))
-    core_attn_out = self.ssm_norm(core.cast(x.dtype) if self.config.ssm and self.config.ssm.kda else core)
-    gate = out_gate.reshape(B, T, self.num_v_heads, self.head_v_dim)
-    gate = gate.float().sigmoid().cast(core_attn_out.dtype) if hasattr(self, "ssm_g_a") else gate.silu()
-    out = (core_attn_out * gate).reshape(B, T, -1)
-    out = out.cast(x.dtype)
-    ret = bf16_partial_linear(out, self.ssm_out.weight).sum(3).cast(dtypes.bfloat16) if resolve(T == 1) and \
-      isinstance(self.ssm_out.weight.device, tuple) and self.ssm_out.weight.shape[1] % (32*len(self.ssm_out.weight.device)) == 0 and \
-      amd_exact_bf16_custom_kernels_supported(out.device) else self.ssm_out(out)
-    return ret.realize(*state_updates)
+    # store the updated state
+    conv_state_store = self.conv_state.uop.store(conv_window[:, 1:, :].cast(self.conv_state.dtype).uop)
+    recurrent_state_store = self.recurrent_state.uop.store(recurrent_state.cast(self.recurrent_state.dtype).uop)
+    recurrent_state = Tensor(self.recurrent_state.uop.after(recurrent_state_store, conv_state_store))
 
-  # Recurrent state can be reused only when the new prompt exactly extends all currently valid state.
-  def _state_tensors(self) -> tuple[Tensor, ...]:
-    if hasattr(self, "conv_state_q"):
-      return self.conv_state_q, self.conv_state_k, self.conv_state_v, self.recurrent_state
-    return (self.conv_state, self.recurrent_state) if hasattr(self, "conv_state") else ()
-  def _state_reset_ops(self): return [s.assign(s.const_like(0)) for s in self._state_tensors()]
-  def _reusable_prefix_len(self, prefix_len:int, cached_len:int) -> int: return prefix_len if prefix_len == cached_len else 0
+    # output
+    core_attn_out = self.ssm_norm((recurrent_state@q).squeeze(-1).reshape(B, 1, self.num_v_heads, self.head_v_dim))
+    out_gate = out_gate.sigmoid() if hasattr(self, "ssm_g_a") else out_gate.silu()
+    return self.ssm_out((core_attn_out * out_gate).reshape(B, 1, -1).cast(x.dtype))
+
+  # recurrent state can't be partially reused after divergence, force a full rebuild
+  def _state_reset_ops(self):
+    return [self.conv_state.assign(self.conv_state.const_like(0)),
+            self.recurrent_state.assign(self.recurrent_state.const_like(0))] if hasattr(self, "conv_state") else []
+  def _reusable_prefix_len(self, prefix_len:int, cached_len:int) -> int: return 0 if prefix_len != cached_len else prefix_len
 
   def _init_state(self, x):
-    if not hasattr(self, "conv_state") and not hasattr(self, "conv_state_q"):
-      if hasattr(self, "attn_q"):
-        device = x.device[0] if isinstance(x.device, tuple) else x.device
-        self.conv_state_q = Tensor.zeros(x.shape[0], self.ssm_conv_kernel-1, self.q_dim, device=device, dtype=x.dtype).clone()
-        self.conv_state_k = Tensor.zeros(x.shape[0], self.ssm_conv_kernel-1, self.q_dim, device=device, dtype=x.dtype).clone()
-        self.conv_state_v = Tensor.zeros(x.shape[0], self.ssm_conv_kernel-1, self.num_v_heads*self.head_v_dim, device=device, dtype=x.dtype).clone()
-        self.recurrent_state = Tensor.zeros(x.shape[0], self.num_v_heads, self.head_v_dim, self.head_k_dim, device=device).clone()
-        if isinstance(x.device, tuple):
-          for state in (self.conv_state_q, self.conv_state_k, self.conv_state_v): state.shard_(x.device, axis=2)
-          self.recurrent_state.shard_(x.device, axis=1)
-      else:
-        self.conv_state = Tensor.zeros(x.shape[0], self.ssm_conv_kernel-1, self.conv_channels, device=x.device, dtype=x.dtype).clone()
-        self.recurrent_state = Tensor.zeros(x.shape[0], self.num_v_heads, self.head_v_dim, self.head_k_dim, device=x.device).clone()
+    if not hasattr(self, "conv_state"):
+      self.conv_state = Tensor.zeros(x.shape[0], self.ssm_conv_kernel-1, self.conv_channels, device=x.device).clone()
+      self.recurrent_state = Tensor.zeros(x.shape[0], self.num_v_heads, self.head_v_dim, self.head_k_dim, device=x.device).clone()
 
 class Transformer:
   def __init__(self, config:TransformerConfig):
-    self.config = config
     dense_config = replace(config, num_experts=0, num_experts_per_tok=0, shared_expert_dim=0, hidden_dim=config.dense_hidden_dim or config.hidden_dim)
     if config.ssm: config = replace(config, qk_norm=config.head_dim)
     block_cls = MLATransformerBlock if config.kv_lora_rank > 0 else TransformerBlock
@@ -558,76 +316,22 @@ class Transformer:
     self.token_embd  = nn.Embedding(config.vocab_size, config.dim)
     self.output_norm = nn.RMSNorm(config.dim, config.norm_eps)
     self.output = Linear(config.dim, config.vocab_size, bias=False)
-    if config.attn_res_block_size:
-      self.output_attn_res_norm = nn.RMSNorm(config.dim, config.norm_eps)
-      self.output_attn_res_proj = Linear(config.dim, 1, bias=False)
     self.max_context = config.max_context
     self.has_recurrent_block = any(isinstance(b, GatedDeltaNetBlock) for b in self.blk)
     self._cached_tokens: list[int] = []
-    self._snapshot_tokens: list[int] = []
-    self._state_snapshots:list[Tensor] = []
-    self._token_buffer:Tensor|None = None
-    self._temperature_buffer:Tensor|None = None
     # we specialize the JIT for prefill and rollout
     self.prefill_jit = TinyJit(self.forward)
     self.rollout_jit = TinyJit(self.forward)
-    self.greedy_prefill_jit = TinyJit(self.forward)
-    self.greedy_rollout_jit = TinyJit(self.forward)
-    self.recurrent_prefill_jits:dict[int, Callable[..., Tensor]] = {}
-    self.recurrent_greedy_prefill_jits:dict[int, Callable[..., Tensor]] = {}
-    self.reset_jit = TinyJit(self._reset_state)
-    self.save_state_jit = TinyJit(self._save_state)
-    self.restore_state_jit = TinyJit(self._restore_state)
 
-  def _reset_state(self) -> None:
-    if resets := [r for b in self.blk for r in b._state_reset_ops()]: Tensor.realize(*resets)
-
-  def _state_tensors(self) -> list[Tensor]:
-    return [s for block in self.blk if isinstance(block, GatedDeltaNetBlock) for s in block._state_tensors()]
-
-  def _init_state_snapshots(self) -> None:
-    if not self._state_snapshots: self._state_snapshots = [s.clone().realize() for s in self._state_tensors()]
-
-  def _save_state(self) -> None:
-    if writes := [dst.assign(src) for dst,src in zip(self._state_snapshots, self._state_tensors())]: Tensor.realize(*writes)
-
-  def _restore_state(self) -> None:
-    if writes := [dst.assign(src) for dst,src in zip(self._state_tensors(), self._state_snapshots)]: Tensor.realize(*writes)
-
-  def forward(self, tokens:Tensor, start_pos:int|UOp, temperature:Tensor|None) -> Tensor:
-    if len(tokens.shape) == 1: tokens = tokens.reshape(1, -1)
-    x = self.token_embd(tokens).cast(dtypes.bfloat16) if self.config.bf16_activations else self.token_embd(tokens).float()
-    block_residual = Tensor.zeros(x.shape[0]*x.shape[1], 0, x.shape[2], device=x.device, dtype=x.dtype) \
-      if self.config.attn_res_block_size else None
-    for i, block in enumerate(self.blk):
-      if block_residual is not None: x, block_residual = block.attn_residual(x, start_pos, block_residual, i)
-      else: x = block(x, start_pos)
-      # Tensor indexing lowers selected experts through a fused one-hot reduction. Keeping all 26
-      # of those high-level graphs alive until the final output is scheduled exhausts host memory.
-      # A realization boundary lowers one block at a time; TinyJit still captures and memory-plans
-      # the resulting schedules for rollout replay.
-      if self.config.expert_mxfp4: x.realize()
-    if block_residual is not None:
-      x = FFNBlock._apply_attn_res(x.reshape(-1, x.shape[-1]), block_residual,
-                                   self.output_attn_res_proj, self.output_attn_res_norm).reshape(x.shape)
-    final_x = self.output_norm(x)
-    if temperature is None and resolve(tokens.numel() == 1) and amd_exact_bf16_custom_kernels_supported(x.device):
-      return bf16_matvec(final_x, self.output.weight).argmax(-1, keepdim=True)
-    logits = self.output(final_x)[:, -1, :]
-    if temperature is None: return logits.argmax(-1, keepdim=True)
+  def forward(self, tokens:Tensor, start_pos:int|UOp, temperature:Tensor) -> Tensor:
+    x = self.token_embd(tokens).float()                   # (B, T, D)
+    for block in self.blk: x = block(x, start_pos)
+    logits = self.output(self.output_norm(x))[:, -1, :]
     # Gumbel-max trick: argmax(logits/temp - log(-log(uniform))) is equivalent to sampling from softmax(logits/temp)
     return (logits / temperature.maximum(1e-12) - (Tensor.rand_like(logits).maximum(1e-12).log().neg()).log()).argmax(-1, keepdim=True)
 
-  def __call__(self, tokens:Tensor, start_pos:int|UOp, temperature:Tensor|None) -> Tensor:
-    token_count = tokens.numel()
-    if self.has_recurrent_block and resolve(token_count != 1):
-      assert isinstance(token_count, int)
-      cache = self.recurrent_greedy_prefill_jits if temperature is None else self.recurrent_prefill_jits
-      jit = cache.setdefault(token_count, TinyJit(self.forward))
-      return jit(tokens.flatten().contiguous(), start_pos, temperature)
-    if temperature is None:
-      return (self.greedy_prefill_jit if resolve(token_count != 1) else self.greedy_rollout_jit)(tokens.flatten().contiguous(), start_pos, None)
-    return (self.prefill_jit if resolve(token_count != 1) else self.rollout_jit)(tokens.flatten().contiguous(), start_pos, temperature)
+  def __call__(self, tokens:Tensor, start_pos:int|UOp, temperature:Tensor) -> Tensor:
+    return (self.prefill_jit if resolve(tokens.shape[1] != 1) else self.rollout_jit)(tokens.contiguous(), start_pos, temperature)
 
   @staticmethod
   def from_gguf(gguf:Tensor|str|pathlib.Path, max_context:int|None=None,
@@ -713,91 +417,31 @@ class Transformer:
     return model, kv
 
   def warmup(self):
-    # Capture the only two shapes used by recurrent serving: a full prefill chunk and one-token rollout.
-    # Two chunks exercise both the initial and nonzero-position prefill paths before the server opens.
-    recurrent_chunk = self.config.recurrent_prefill_chunk_size or 32
-    prompt = [0] * max(1, min(recurrent_chunk*2, self.max_context-2)) if self.has_recurrent_block else [0]
-    # Recurrent serving captures both greedy and sampled graphs, then executes one replay so graph
-    # creation/lowering cannot leak into request latency for either HTTP temperature path.
-    # generate mutates its token list, so each pass needs a fresh prompt to exercise cache reset.
-    for temperature in ((0.0, 1.0) if self.has_recurrent_block else (0.0,)):
-      for _ in range(3 if self.has_recurrent_block else 2): list(zip(range(2), self.generate(prompt.copy(), temperature=temperature)))
-    # Capture prompt-boundary restore using successively extended prompts so every restore starts
-    # from the checkpoint made by the previous pass.
-    if self.has_recurrent_block:
-      for i in range(1, 4): list(zip(range(2), self.generate(prompt + list(range(1, i+1)), temperature=0.0)))
+    for _ in range(2): list(zip(range(2), self.generate([0])))
 
-  def _cache_start(self, tokens:list[int]) -> tuple[int, bool]:
+  def get_start_pos(self, tokens:list[int]) -> int:
     prefix_len = sum(1 for _ in itertools.takewhile(lambda ab: ab[0] == ab[1], zip(tokens[:-1], self._cached_tokens)))
-    live_start = min(block._reusable_prefix_len(prefix_len, len(self._cached_tokens)) for block in self.blk)
-    snapshot_prefix = sum(1 for _ in itertools.takewhile(lambda ab: ab[0] == ab[1], zip(tokens[:-1], self._snapshot_tokens)))
-    snapshot_start = len(self._snapshot_tokens) if snapshot_prefix == len(self._snapshot_tokens) else 0
-    return (snapshot_start, True) if snapshot_start > live_start else (live_start, False)
-
-  def get_start_pos(self, tokens:list[int]) -> int: return self._cache_start(tokens)[0]
+    return min(block._reusable_prefix_len(prefix_len, len(self._cached_tokens)) for block in self.blk)
 
   def generate(self, tokens:list[int], chunk_size:int=32, temperature:float=0.0):
-    chunked_recurrent = self.has_recurrent_block and self.config.recurrent_prefill_chunked
-    if chunked_recurrent and self.config.recurrent_prefill_chunk_size:
-      chunk_size = min(chunk_size, self.config.recurrent_prefill_chunk_size)
-    if self.has_recurrent_block and not chunked_recurrent: chunk_size = 1
+    if self.has_recurrent_block: chunk_size = 1
     v_start_pos = UOp.variable("start_pos", 0, self.max_context-1)
     v_toks = UOp.variable("toks", 1, chunk_size)
     # TODO: use UOp.variable for temperature once float variables are supported
-    model_device = self.token_embd.weight.device
-    amd_tp = isinstance(model_device, tuple) and all(d.startswith("AMD") for d in model_device)
-    if temperature == 0.0: temp = None
-    elif amd_tp:
-      if self._temperature_buffer is None: self._temperature_buffer = Tensor.empty(1, device=model_device).realize()
-      temp_storage = self._temperature_buffer.uop.buf_uop.buffer
-      temp_buffers = temp_storage.bufs if isinstance(temp_storage, MultiBuffer) else [temp_storage]
-      temp_host = memoryview(array.array('f', [temperature])).cast('B')
-      for buf in temp_buffers: buf.ensure_allocated().allocator._copyin(buf._buf, temp_host)
-      temp = self._temperature_buffer
-    else: temp = Tensor([temperature], device=model_device)
-    # Keep the replicated AMD token buffer identity stable across HTTP requests so captured graphs
-    # see the same input topology. Updating this small int32 buffer is cheaper than rebuilding JITs.
-    if amd_tp:
-      if self._token_buffer is None:
-        self._token_buffer = Tensor.empty(1, self.max_context, dtype=dtypes.int32, device=model_device).realize()
-      token_storage = self._token_buffer.uop.buf_uop.buffer
-      token_buffers = token_storage.bufs if isinstance(token_storage, MultiBuffer) else [token_storage]
-      input_host = memoryview(array.array('i', tokens + [0] * (self.max_context-len(tokens)))).cast('B')
-      for buf in token_buffers: buf.ensure_allocated().allocator._copyin(buf._buf, input_host)
-      t = self._token_buffer
-    else: t = Tensor(tokens + [0] * (self.max_context - len(tokens)), dtype="int32", device=model_device).reshape(1, self.max_context)
+    temp = Tensor([temperature])
+    # assign all input tokens once, then slice from start_pos for the model call
+    t = Tensor(tokens + [0] * (self.max_context - len(tokens)), dtype="int32").reshape(1, self.max_context)
     # recompute start_pos from what's currently valid in the caches
-    start_pos, restore_snapshot = self._cache_start(tokens)
-    # This graph is captured by warmup. Resetting on-device avoids hundreds of synchronous copies
-    # to sharded AMD state buffers and guarantees unrelated requests don't schedule new kernels.
-    if restore_snapshot: self.restore_state_jit()
-    elif start_pos < len(self._cached_tokens) and self.has_recurrent_block: self.reset_jit()
+    start_pos = self.get_start_pos(tokens)
+    if start_pos < len(self._cached_tokens) and (resets := [r for b in self.blk for r in b._state_reset_ops()]): Tensor.realize(*resets)
     out, prompt_len = None, len(tokens)
-    token_host = memoryview(bytearray(4)) if amd_tp else None
     while len(tokens) < self.max_context:
-      remaining = len(tokens) - start_pos
-      # Full recurrent chunks use the high-throughput prefill graph. Process the tail through the
-      # rollout graph so every request uses only the two shapes captured during server warmup.
-      n_toks = chunk_size if chunked_recurrent and remaining >= chunk_size else 1 if chunked_recurrent else min(chunk_size, remaining)
-      # Recurrent blocks execute an explicit recurrence over T. Give them a static chunk length so
-      # Python constructs the recurrence once per encountered size; decode remains the T=1 JIT.
-      if chunked_recurrent:
-        # Token count is static for the recurrent kernel, but cache position must remain a runtime
-        # variable so repeated chunks do not replay MLA stores at the capture position.
-        sp = v_start_pos.bind(start_pos)
-        model_input = t[:, sp:sp+n_toks] if start_pos < prompt_len or out is None else out
-      else:
-        sp = v_start_pos.bind(start_pos)
-        nt = v_toks.bind(n_toks)
-        model_input = t[:, sp:sp+nt] if start_pos < prompt_len or out is None else out
-      out = self(model_input, sp, temp).realize()
+      n_toks = min(chunk_size, len(tokens) - start_pos)
+      sp, nt = v_start_pos.bind(start_pos), v_toks.bind(n_toks)
+      out = self(t[:, sp:sp+nt] if start_pos < prompt_len or out is None else out, sp, temp).realize()
       start_pos += n_toks
       # chunked prefill: keep processing until all prompt tokens are consumed
       if start_pos < len(tokens): continue
-      if self.has_recurrent_block and len(tokens) == prompt_len and self._state_tensors():
-        self._init_state_snapshots()
-        self.save_state_jit()
-        self._snapshot_tokens = tokens.copy()
-      tokens.append(amd_int32_item(out, token_host) if token_host is not None else int(out.item()))
+      tokens.append(int(out.item()))
       self._cached_tokens = tokens[:-1]
       yield tokens[-1]

tinygrad/llm/quant.py

@@ -1,81 +0,0 @@
-from tinygrad import Tensor, dtypes
-
-MX_BLOCK_SIZE = 32
-MXFP4_VALUES = (0.0, 0.5, 1.0, 1.5, 2.0, 3.0, 4.0, 6.0,
-                -0.0, -0.5, -1.0, -1.5, -2.0, -3.0, -4.0, -6.0)
-
-def _e8m0_scale(scale:Tensor) -> Tensor:
-  """Decode an OCP E8M0 scale byte. 127 encodes 2**0."""
-  return (scale.cast(dtypes.float32) - 127.0).exp2()
-
-def quantize_mxfp4(x:Tensor) -> tuple[Tensor, Tensor]:
-  """Quantize the last dimension to OCP MXFP4 (E2M1 values, E8M0 scale, block size 32)."""
-  if x.shape[-1] % MX_BLOCK_SIZE: raise ValueError(f"MXFP4 requires a multiple-of-32 last dimension, got {x.shape}")
-  *outer, k = x.shape
-  blocks = x.float().reshape(*outer, k//MX_BLOCK_SIZE, MX_BLOCK_SIZE)
-  amax = blocks.abs().max(axis=-1)
-  # Match the OCP/MLX reference: quantize amax / E2M1_MAX to the nearest E8M0
-  # power of two. This deliberately differs from extracting exponent bits: blocks
-  # whose maximum is near a power-of-two boundary can select a scale 2x smaller.
-  exponent = (amax.maximum(1e-38).div(6.0).log2().round()).clamp(-127, 127)
-  scale = (amax == 0).where(127, exponent + 127).cast(dtypes.uint8)
-  normalized = blocks / _e8m0_scale(scale).unsqueeze(-1)
-  # Midpoint bins avoid materializing a 16x larger distance tensor while preserving nearest-value encoding.
-  magnitude = normalized.abs()
-  # OCP formats use round-to-nearest-even: at alternating midpoints, the upper code has an even mantissa LSB.
-  code = sum(((magnitude >= midpoint) if upper_even else (magnitude > midpoint)).cast(dtypes.uint8)
-             for midpoint, upper_even in ((0.25, False), (0.75, True), (1.25, False), (1.75, True),
-                                           (2.5, False), (3.5, True), (5.0, False)))
-  code = ((normalized < 0) & (code != 0)).where(code + 8, code).reshape(*outer, k)
-  # Safetensors has no nibble dtype. Store the earlier element in the low nibble.
-  packed = code[..., ::2] + code[..., 1::2] * 16
-  return packed.contiguous(), scale.contiguous()
-
-def quantize_mxfp4_cpu(x:Tensor) -> tuple[Tensor, Tensor]:
-  """CPU converter fast path for large checkpoints. Inference itself does not depend on numpy."""
-  import numpy as np
-  if x.shape[-1] % MX_BLOCK_SIZE: raise ValueError(f"MXFP4 requires a multiple-of-32 last dimension, got {x.shape}")
-  array = x.float().numpy()
-  blocks = array.reshape(*array.shape[:-1], array.shape[-1]//MX_BLOCK_SIZE, MX_BLOCK_SIZE)
-  amax = np.max(np.abs(blocks), axis=-1)
-  exponent = np.clip(np.rint(np.log2(np.maximum(amax, 1e-38) / 6.0)), -127, 127)
-  scale = np.where(amax == 0, 127, exponent + 127).astype(np.uint8)
-  normalized = blocks / np.exp2(scale.astype(np.float32) - 127)[..., None]
-  magnitude = np.abs(normalized)
-  code = sum(((magnitude >= midpoint) if upper_even else (magnitude > midpoint)).astype(np.uint8)
-             for midpoint, upper_even in ((0.25, False), (0.75, True), (1.25, False), (1.75, True),
-                                           (2.5, False), (3.5, True), (5.0, False)))
-  code = np.where((normalized < 0) & (code != 0), code + 8, code).astype(np.uint8).reshape(array.shape)
-  packed = code[..., ::2] + code[..., 1::2] * 16
-  return Tensor(packed), Tensor(scale)
-
-def dequantize_mxfp4(packed:Tensor, scale:Tensor, dtype=dtypes.bfloat16) -> Tensor:
-  """Decode the packed representation emitted by quantize_mxfp4."""
-  if packed.shape[-1] != scale.shape[-1] * 16:
-    raise ValueError(f"incompatible MXFP4 values/scales: {packed.shape} and {scale.shape}")
-  lo = packed - packed.div(16, rounding_mode="trunc") * 16
-  hi = packed.div(16, rounding_mode="trunc")
-  code = Tensor.stack(lo, hi, dim=-1).reshape(*packed.shape[:-1], packed.shape[-1]*2)
-  values = Tensor(MXFP4_VALUES, dtype=dtypes.float32, device=packed.device)[code]
-  scales = _e8m0_scale(scale).unsqueeze(-1).expand(*scale.shape, MX_BLOCK_SIZE).reshape(*scale.shape[:-1], scale.shape[-1]*MX_BLOCK_SIZE)
-  return (values * scales).cast(dtype)
-
-def quantize_dequantize_mxfp8(x:Tensor, dtype=dtypes.bfloat16) -> Tensor:
-  """Apply the Kimi expert-activation MXFP8 E4M3/E8M0 round trip in 32-value blocks."""
-  if x.shape[-1] % MX_BLOCK_SIZE: raise ValueError(f"MXFP8 requires a multiple-of-32 last dimension, got {x.shape}")
-  *outer, k = x.shape
-  blocks = x.float().reshape(*outer, k//MX_BLOCK_SIZE, MX_BLOCK_SIZE)
-  amax = blocks.abs().max(axis=-1)
-  # As for MXFP4, the E8M0 scale is nearest-power-of-two(amax / E4M3_MAX).
-  exponent = (amax.maximum(1e-38).div(448.0).log2().round()).clamp(-127, 127)
-  scale = (amax == 0).where(127, exponent + 127).cast(dtypes.uint8)
-  normalized = blocks / _e8m0_scale(scale).unsqueeze(-1)
-  # Software OCP E4M3 rounding is required on gfx1100 (RDNA3 has no native FP8 dtype).
-  # E4M3 has three explicit mantissa bits and a minimum normal exponent of -6;
-  # using e=-6 also gives the 2**-9 subnormal quantum.
-  magnitude = normalized.abs().clamp(max_=448.0)
-  elem_exp = magnitude.maximum(2**-9).log2().floor().clamp(-6, 8)
-  quantum = (elem_exp - 3).exp2()
-  quantized = (magnitude / quantum).round() * quantum
-  quantized = (normalized < 0).where(-quantized, quantized).clamp(-448.0, 448.0)
-  return (quantized * _e8m0_scale(scale).unsqueeze(-1)).reshape(*outer, k).cast(dtype)

tinygrad/llm/serve.py

@@ -24,11 +24,6 @@ def parse_tool_call(s:str) -> tuple[str, typing.Any]|None:
     return fm.group(1), args
   return None
 
-def parse_kimi_tool_call(s:str) -> tuple[str, typing.Any]|None:
-  if (m := re.match(r"\s*(?:functions\.)?([^:\s]+)(?::[^\s]+)?\s*<\|tool_call_argument_begin\|>\s*(.*?)\s*\Z", s, re.DOTALL)) is None: return None
-  try: return m.group(1), json.loads(m.group(2))
-  except json.JSONDecodeError: return None
-
 def normalize_messages(messages:list[dict]) -> None:
   # chat templates expect tool_call arguments as dicts (OpenAI clients send JSON strings)
   for m in messages:
@@ -39,10 +34,9 @@ def normalize_messages(messages:list[dict]) -> None:
 
 class StreamRouter:
   # routes streamed output text to (field, text) deltas, keeping tool_call regions in .buf for the final parse
-  def __init__(self, reasoning:bool=False, xtml:bool=False):
+  def __init__(self, reasoning:bool=False):
     self.buf = ""
     self.mode = "reasoning" if reasoning else "undecided"  # output inside a think block is sent as reasoning_content
-    self.xtml = xtml
   def split(self, tag:str, final:bool) -> tuple[str, bool]:
     # split buf on the first full tag, holding back a partial tag at the end unless final
     if tag in self.buf:
@@ -51,39 +45,20 @@ class StreamRouter:
     hold = max((i for i in range(1, min(len(self.buf), len(tag))+1) if tag.startswith(self.buf[-i:])), default=0) if not final else 0
     emit, self.buf = self.buf[:len(self.buf)-hold], self.buf[len(self.buf)-hold:]
     return emit, False
-  def split_any(self, tags:tuple[str, ...], final:bool) -> tuple[str, str|None]:
-    found = [(self.buf.index(tag), tag) for tag in tags if tag in self.buf]
-    if found:
-      pos, tag = min(found)
-      before, self.buf = self.buf[:pos], self.buf[pos+len(tag):]
-      return before, tag
-    hold = max((i for tag in tags for i in range(1, min(len(self.buf), len(tag))+1) if tag.startswith(self.buf[-i:])), default=0) if not final else 0
-    emit, self.buf = self.buf[:len(self.buf)-hold], self.buf[len(self.buf)-hold:]
-    return emit, None
   def route(self, piece:str, final:bool=False) -> typing.Iterator[tuple[str, str]]:
     self.buf += piece
     if self.mode == "undecided":  # decide whether the output starts with a think block
       if not final and len(self.buf) < len("<think>") and "<think>".startswith(self.buf): return
       self.mode, self.buf = ("reasoning", self.buf[len("<think>"):]) if self.buf.startswith("<think>") else ("content", self.buf)
     if self.mode == "reasoning":
-      emit, done = self.split("<|close|>think<|sep|>" if self.xtml else "</think>", final)
+      emit, done = self.split("</think>", final)
       if emit: yield "reasoning_content", emit
       if not done: return
-      self.mode = "content_open" if self.xtml else "content"
-    if self.mode == "content_open":
-      _, found = self.split("<|open|>response<|sep|>", final)
-      if not found: return
       self.mode = "content"
-    if self.mode == "done": return
-    if self.xtml and self.mode == "content":
-      emit, found = self.split("<|close|>response<|sep|>", final)
-      if emit: yield "content", emit
-      if found: self.mode = "done"
-      return
     if self.mode == "tool": return
-    emit, tool_tag = self.split_any(("<tool_call>", "<|tool_calls_section_begin|>"), final)
+    emit, found = self.split("<tool_call>", final)
     if emit: yield "content", emit
-    if tool_tag: self.mode, self.buf = "tool", tool_tag + self.buf
+    if found: self.mode, self.buf = "tool", "<tool_call>" + self.buf
 
 class Handler(HTTPRequestHandler):
   server: LLMServer
@@ -92,7 +67,7 @@ class Handler(HTTPRequestHandler):
     if self.path == "/v1/models": self.send_data(json.dumps({"object":"list","data":[{"id":self.server.model_name,"object":"model"}]}).encode())
     else: self.send_data((pathlib.Path(__file__).parent / "chat.html").read_bytes(), content_type="text/html")
   def run_model(self, ids:list[int], model_name:str, include_usage=False, max_tokens:int|None=None, temperature:float=0.0,
-                reasoning:bool=False, xtml:bool=False):
+                reasoning:bool=False):
     model, tok = self.server.model, self.server.tok
     prompt_tokens = len(ids)
     cache_start_pos = model.get_start_pos(ids)
@@ -103,7 +78,7 @@ class Handler(HTTPRequestHandler):
     finish_reason = "stop"
     st = pt = time.perf_counter()
     dec = tok.stream_decoder()
-    router = StreamRouter(reasoning, xtml)
+    router = StreamRouter(reasoning)
     def log_stats(interrupted:bool=False):
       et = time.perf_counter()
       total = f"total:{et-st:6.2f}s"
@@ -131,14 +106,6 @@ class Handler(HTTPRequestHandler):
           name, args = parsed
           tool_calls.append({"index":len(tool_calls), "id":f"call_{uuid.uuid4().hex[:24]}", "type":"function",
                              "function":{"name":name, "arguments":args if isinstance(args, str) else json.dumps(args)}})
-      for m in re.finditer(r"<\|tool_call_begin\|>(.*?)<\|tool_call_end\|>", router.buf, re.DOTALL):
-        if (parsed := parse_kimi_tool_call(m.group(1))) is None:
-          stderr_log(f"failed to parse Kimi tool call: {m.group(1)[:200]}")
-          yield chunk({"content":m.group(0)})
-        else:
-          name, args = parsed
-          tool_calls.append({"index":len(tool_calls), "id":f"call_{uuid.uuid4().hex[:24]}", "type":"function",
-                             "function":{"name":name, "arguments":json.dumps(args)}})
       if tool_calls:
         yield chunk({"tool_calls":tool_calls})
         if finish_reason == "stop": finish_reason = "tool_calls"
@@ -172,10 +139,9 @@ class Handler(HTTPRequestHandler):
 
       # reply
       max_tokens = body.get("max_completion_tokens") or body.get("max_tokens")
-      xtml = rendered.rstrip().endswith("<|open|>think<|sep|>")
       chunks = self.run_model(ids, body["model"], not body.get("stream") or body.get("stream_options",{}).get("include_usage", False),
                               max_tokens=max_tokens, temperature=float(body.get("temperature", 0.0)),
-                              reasoning=xtml or rendered.rstrip().endswith("<think>"), xtml=xtml)
+                              reasoning=rendered.rstrip().endswith("<think>"))
       if body.get("stream"): self.stream_json(chunks)
       else:
         out, reasoning, tool_calls, finish_reason = [], [], [], "stop"

tinygrad/nn/state.py

@@ -1,4 +1,4 @@
-import json, math, pathlib, zipfile, pickle, tarfile, struct, functools, io, zlib
+import json, math, pathlib, struct, functools, io, zlib
 from collections import OrderedDict
 from typing import Any, Callable, BinaryIO, Iterable, cast
 from tinygrad.tensor import Tensor
@@ -219,6 +219,7 @@ def load_state_dict(model, state_dict:dict[str, Tensor], strict=True, verbose=Tr
 
 @accept_filename
 def zip_extract(t: Tensor) -> dict[str, Tensor]:
+  import zipfile
   files: dict[str, Tensor] = {}
   with zipfile.ZipFile(TensorIO(t), "r") as myzip:
     # sadly, the extra length needs to be read from the local header of each file.
@@ -249,6 +250,7 @@ def tar_extract(t: Tensor) -> dict[str, Tensor]:
   tensors = nn.state.tar_extract(Tensor(pathlib.Path("archive.tar")))
   ```
   """
+  import tarfile
   with tarfile.open(fileobj=TensorIO(t), mode="r") as tar:
     return {member.name:t[member.offset_data:member.offset_data+member.size] for member in tar if member.type == tarfile.REGTYPE}
 
@@ -303,6 +305,7 @@ def torch_load(t:Tensor) -> dict[str, Tensor]:
                "FloatTensor": None, "Parameter": Parameter}
   whitelist = {"torch", "collections", "numpy", "_codecs"}  # NOTE: this is not for security, only speed
   class Dummy: pass
+  import pickle, zipfile, tarfile
   class TorchPickle(pickle.Unpickler):
     def find_class(self, module, name):
       module_root = module.split(".")[0]

tinygrad/renderer/llvmir.py

@@ -279,43 +279,9 @@ exit: %packed = phi i32 [%packed_bf8, %do_bf8], [%packed_fp8, %do_fp8]\n  %trunc
       (UPat(Ops.WMMA, name="wmma"), lambda ctx, wmma, rdna4=AMDLLVMRenderer.is_rdna4(target.arch), cdna=self.is_cdna:
         render_wmma_amd(ctx, wmma, cdna, rdna4))
     ])
-    if self.is_cdna:
-      self.extra_matcher += PatternMatcher([
-        (UPat(Ops.WMMA, name="x", dtype=dtypes.float),
-          lambda x: x.replace(src=(x.src[0].bitcast(dtypes.uint32), x.src[1].bitcast(dtypes.uint32), x.src[2]))
-          if x.arg[0][2] == 128 and x.src[0].dtype.itemsize <= 8 else None),
-        (UPat(Ops.WMMA, name="x", dtype=dtypes.float),
-          lambda x: x.replace(src=(x.src[0].bitcast(dtypes.uint16), x.src[1].bitcast(dtypes.uint16), x.src[2]))
-          if x.max_numel() == 4 and x.src[0].dtype == dtypes.bfloat16 and x.src[0].max_numel() == 4 else None),
-        (UPat(Ops.WMMA, name="x", dtype=dtypes.float),
-          lambda x: x.replace(src=(x.src[0].bitcast(dtypes.uint64), x.src[1].bitcast(dtypes.uint64), x.src[2]))
-          if x.max_numel() == 4 and x.src[0].dtype in dtypes.fp8_ocp and x.src[0].max_numel() == 8 else None),
-      ])
-    if target.arch in {"gfx1100", "gfx1151"}:
-      self.extra_matcher += PatternMatcher([
-        (UPat(Ops.WMMA, name="x", dtype=dtypes.int32), lambda x: x.replace(
-          src=(x.src[0].bitcast(dtypes.uint32), x.src[1].bitcast(dtypes.uint32), x.src[2]))
-          if x.src[0].dtype == dtypes.int8 and x.src[0].max_numel() == 16 else None),
-        (UPat(Ops.WMMA, name="x", dtype=dtypes.half), lambda x: UOp(Ops.STACK, src=tuple(x.replace(
-          src=(x.src[0], x.src[1], UOp(Ops.STACK, src=tuple(x.src[2].index(UOp.const(j//2, dtypes.int16))
-            if j%2 == 0 else UOp.const(0.0, x.src[2].dtype)
-            for j in range(x.max_numel()*2)))),
-          arg=(*x.arg[:4], None)).index(UOp.const(i*2, dtypes.int16))
-          for i in range(x.max_numel()))) if x.max_numel() == 8 else None),
-        (UPat(Ops.WMMA, name="x"), lambda x: x.replace(
-          src=(x.src[0].bitcast(dtypes.uint16), x.src[1].bitcast(dtypes.uint16), x.src[2]))
-          if x.src[0].dtype == dtypes.bfloat16 and x.src[0].max_numel() == 16 else None),
-      ])
-    if target.arch in {"gfx1200", "gfx1201"}:
-      self.extra_matcher += PatternMatcher([
-        (UPat(Ops.WMMA, name="x", dtype=dtypes.bfloat16), lambda x: x.replace(
-          dtype=dtypes.uint16,
-          src=(x.src[0].bitcast(dtypes.uint16), x.src[1].bitcast(dtypes.uint16), x.src[2].bitcast(dtypes.uint16)))
-            .bitcast(dtypes.bfloat16) if x.max_numel() == 8 and x.src[0].dtype == dtypes.bfloat16 and x.src[0].max_numel() == 8 else None),
-        (UPat(Ops.WMMA, name="x", dtype=dtypes.float),
-          lambda x: x.replace(src=(x.src[0].bitcast(dtypes.uint16), x.src[1].bitcast(dtypes.uint16), x.src[2]))
-          if x.max_numel() == 8 and x.src[0].dtype == dtypes.bfloat16 and x.src[0].max_numel() == 8 else None)
-      ])
+    if self.is_cdna: self.extra_matcher += tc.pm_validate_wmma_cdna
+    if target.arch in {"gfx1100", "gfx1151"}: self.extra_matcher += tc.pm_validate_wmma_rdna3
+    if target.arch in {"gfx1200", "gfx1201"}: self.extra_matcher += tc.pm_validate_wmma_rdna4
 
   def supported_dtypes(self): return {d for d in super().supported_dtypes()
                                       if (d not in dtypes.fp8_ocp or self.target.arch == "gfx950") and d not in dtypes.fp8_fnuz}

tinygrad/runtime/graph/metal.py

@@ -113,5 +113,6 @@ class MetalGraph(GraphRunner):
   @staticmethod
   def supports_uop(batch_devs, new_call:UOp) -> bool:
     # Metal ICB replay encodes offsets as uint32; reject if any Metal buffer offset exceeds 32-bit range.
-    if any(b.op is Ops.SLICE and b.src[1].val * b.src[0].dtype.itemsize > 0xFFFFFFFF for b in new_call.src[1:]): return False
+    for shrink in [s for src in new_call.src[1:] if (s:=src.src[0] if src.op is Ops.BITCAST else src).op is Ops.SHRINK]:
+      if shrink.src[1].val * shrink.src[0].dtype.itemsize > 0xFFFFFFFF: return False
     return GraphRunner.supports_uop(batch_devs, new_call)

tinygrad/runtime/support/hcq2.py

@@ -33,13 +33,13 @@ class HCQInfo:
 
   input_idxs:tuple[int, ...] = () # indexes into input_uops used by this call
   inputs:int|None = None
-  prof:tuple[ProfileGraphEntry, ...] = () # st_id/en_id are timestamp signal slots until collect
+  kernels:tuple[tuple[tuple[str, ...], str, Estimates, tuple[int, ...]], ...] = ()
 
 def all_devices_in(d:Any, c:frozenset[str]) -> bool: return {x.split(":")[0] for x in to_tuple(d)} <= c
 
 def unwrap_mstack(u):
   if u.op is Ops.MSTACK: return tuple(x for s in u.src for x in unwrap_mstack(s))
-  return unwrap_mstack(u.src[0]) if u.op in {Ops.MSELECT, Ops.SLICE} else (u,)
+  return unwrap_mstack(u.src[0]) if u.op is Ops.MSELECT else (u,)
 
 def is_value_known_at_link(val:UOp) -> bool:
   runtime_reads = [u for u in val.toposort() if u.op in (Ops.LOAD, Ops.INDEX)]
@@ -96,12 +96,24 @@ pm_replace_buffers = PatternMatcher([(UPat(Ops.CALL, name="call"), replace_call_
 
 def _need_staging(a, b): return all_devices_in(a.device, HCQ_DEVS) and not all_devices_in(b.device, HCQ_P2P_DEVS)
 
+def hcq_call_devs(call:UOp) -> Any|None: return next((b.device for b in call.src[1:] if all_devices_in(b.device, HCQ_DEVS)), None)
+
 def stage_copy(dst:UOp, src:UOp) -> UOp|None:
   if not (_need_staging(src, dst) or _need_staging(dst, src)): return None
 
   stage = UOp.new_buffer("CPU", src.max_numel() * src.dtype.itemsize, dtypes.uint8)
   return UOp(Ops.LINEAR, src=(src.copy_to_device("CPU").call(stage, src), stage.copy_to_device(dst.device).call(dst, stage)))
-pm_insert_copy_staging = PatternMatcher([(UPat(Ops.CALL, src=(UPat(Ops.COPY), UPat(name="dst"), UPat(name="src"))), stage_copy)])
+
+def kernel_copy(call:UOp, dst:UOp, src:UOp) -> UOp|None:
+  if (devs:=hcq_call_devs(call)) is None or Device[(dev:=to_tuple(devs)[0])].has_copy_queue: return None
+  d, s = (UOp.param(i, dst.dtype, (n:=dst.max_numel(),), device=devs) for i in range(2))
+  ast = d.index(r:=UOp.range(n, 0)).store(s.index(r).load()).end(r).sink(arg=KernelInfo(name="copy"), tag=1)
+  return call.replace(src=(to_program(ast, Device[dev].renderer), dst, src))
+
+pm_insert_copy_staging = PatternMatcher([
+  (UPat(Ops.CALL, src=(UPat(Ops.COPY), UPat(name="dst"), UPat(name="src"))), stage_copy),
+  (UPat(Ops.CALL, src=(UPat(Ops.COPY), UPat(name="dst"), UPat(name="src")), name="call"), kernel_copy)
+])
 
 # *****************
 # 2. deps
@@ -129,14 +141,14 @@ def _build_wait_cmds(slots:dict[str, int], dep_lanes:list[tuple[tuple, int, int]
 
   # opt2: keep latest dep per (dep device, queue, cur lane)
   latest = {((dep[0][dlane], dep[1]), lane): (dep, dlane) for dep, dlane, lane in sorted(dep_lanes, key=lambda x: x[0][2])}
-  deps:dict[tuple, list[int|None]] = collections.defaultdict(lambda: [None]*len(devices))
-  for (_, lane), (dep, dlane) in latest.items(): deps[dep][lane] = dlane
+  deps:dict[tuple, dict[int, list[int]]] = collections.defaultdict(lambda: collections.defaultdict(list))
+  for (_, lane), (dep, dlane) in latest.items(): deps[dep][lane].append(dlane)
 
   waits = []
-  for (ddevs, dqueue, dtag), lanes in deps.items():
-    sig = UOp.mstack(*[make_signal(d, tag="sentinel_signal") if dl is None else make_signal(ddevs[dl], slots[dqueue])
-                       for dl, d in zip(lanes, devices)])
-    waits.append(UOp(Ops.INS, arg="wait", src=(sig, UOp.const(dtag + 1, dtypes.uint64))))
+  for (ddevs, dqueue, dtag), by_lane in deps.items():
+    for ls in itertools.zip_longest(*(by_lane[lane] for lane in range(len(devices)))):
+      s = UOp.mstack(*[make_signal(d, tag="sentinel_signal") if dl is None else make_signal(ddevs[dl], slots[dqueue]) for dl, d in zip(ls, devices)])
+      waits.append(UOp(Ops.INS, arg="wait", src=(s, UOp.const(dtag + 1, dtypes.uint64))))
   return waits, {dtag for _, _, dtag in deps}
 
 def _build_finalizers(batch:list[tuple[UOp, tuple[str, ...]]], batch_info:list[tuple[tuple[str, ...], str]],
@@ -187,10 +199,10 @@ def _finalize_batch(batch:list[tuple[UOp, tuple[str, ...]]], profile:bool) -> li
     signal_tags |= cur_signal_tags
 
   # build fences and finalizers
-  fences, finalizers, finalizer_signal_tags = _build_finalizers(batch, batch_info, deps_tracker, slots)
+  fences, fins, finalizer_signal_tags = _build_finalizers(batch, batch_info, deps_tracker, slots)
   signal_tags |= finalizer_signal_tags
 
-  src, prof = [], []
+  src, kerns = [], []
   for tag, ((call, _), (devices, queue), q) in enumerate(zip(batch, batch_info, call_waits)):
     # first queue use, sync prior device work with the device timeline
     if batch_info.index((devices, queue)) == tag:
@@ -200,24 +212,24 @@ def _finalize_batch(batch:list[tuple[UOp, tuple[str, ...]]], profile:bool) -> li
     # and make hcq call
     name, info = get_call_name(call, get_call_arg_uops(call)), HCQInfo(devices, estimate_uop(call))
     ts_ids = [next(UOp.unique_num) for _ in range(2)] if profile else []
-    prof += [ProfileGraphEntry(d, name, *ts_ids) for d in devices if ts_ids]
+    kerns.append((devices, name, info.estimates, tuple(ts_ids)))
 
     ts_ins = [UOp(Ops.INS, arg="timestamp", src=(make_signal(devices, s),)) for s in ts_ids]
     q += ts_ins[:1] + [call.replace(arg=replace(call.arg, aux=info))] + ts_ins[1:]
 
     # signal the queue if someone waits for us
     if tag in signal_tags: q += [UOp(Ops.INS, arg="store", src=(make_signal(devices, slots[queue]), UOp.const(tag + 1, dtypes.uint64)))]
-    src.append(make_call(name, make_submit(*q, devs=devices, queue=queue).sink(), info))
+    src.append(make_call(f"submit {name}", make_submit(*q, devs=devices, queue=queue).sink(), info))
 
   # append batch timestamps to finalizers
-  finalizers = [f.replace(arg=replace(f.arg, aux=replace(a:=f.arg.aux, prof=tuple(e for e in prof if e.device in a.device)))) for f in finalizers]
-  return fences + src + finalizers
+  fins = [f.replace(arg=replace(f.arg, aux=replace(a:=f.arg.aux, kernels=tuple(x for x in kerns if set(x[0]) & set(a.device))))) for f in fins]
+  return fences + src + fins
 
 def sched_hcq_batches(l:UOp, profile:bool) -> UOp:
   srcs:list[UOp] = []
   batch:list[tuple[UOp, tuple[str, ...]]] = []
   for call in l.src:
-    if (devs:=next((b.device for b in call.src[1:] if all_devices_in(b.device, HCQ_DEVS)), None)) is not None: batch.append((call, to_tuple(devs)))
+    if (devs:=hcq_call_devs(call)) is not None: batch.append((call, to_tuple(devs)))
     else: srcs, batch = srcs + _finalize_batch(batch, profile) + [call], []
   return l.replace(src=tuple(srcs + _finalize_batch(batch, profile)))
 
@@ -365,14 +377,15 @@ pm_replace_params = PatternMatcher([
 
 # *****************
 
-def resolve_getaddr_slice(bv:UOp, g:UOp) -> UOp:
+def resolve_getaddr_view(bv:UOp, g:UOp) -> UOp:
   base = bv.src[0].after(*g.src[0].src[1:] if g.src[0].op is Ops.AFTER else ())
-  itemsize = bv.src[0].dtype.itemsize if bv.src[0].without_after.op in (Ops.BUFFER, Ops.SLICE, Ops.MSTACK, Ops.MSELECT) else bv.dtype.itemsize
+  if bv.op is Ops.BITCAST: return UOp(Ops.GETADDR, src=(base,), arg=g.arg)
+  itemsize = bv.src[0].dtype.itemsize if bv.src[0].without_after.op in (Ops.BUFFER, Ops.MSTACK, Ops.MSELECT) else bv.dtype.itemsize
   return UOp(Ops.GETADDR, src=(base,), arg=g.arg) + UOp.const(bv.src[1].val * itemsize, dtypes.uint64)
 
 pm_early_simplify = PatternMatcher([
-  (UPat(Ops.GETADDR, src=(UPat.any(sl:=UPat(Ops.SLICE, name="bv"), sl.after(allow_any_len=True)),), name="g"), resolve_getaddr_slice),
-  (UPat(Ops.INDEX, src=(UPat(Ops.SLICE, name="bv"),), allow_any_len=True, name="x"),
+  (UPat(Ops.GETADDR, src=(UPat((Ops.SHRINK, Ops.BITCAST), name="bv").or_after(),), name="g"), resolve_getaddr_view),
+  (UPat(Ops.INDEX, src=(UPat(Ops.SHRINK, name="bv"),), allow_any_len=True, name="x"),
    lambda bv,x: x.replace(src=(bv.src[0], x.src[1] + bv.src[1].cast(x.src[1].dtype), *x.src[2:]))),
 ])
 
@@ -390,7 +403,7 @@ def pack_hcq_placeholders(call:UOp) -> UOp|None:
       sizes[b.tag] = offs[b] + b.max_numel()
   counts = collections.Counter(b.tag for b in bufs)
   bases = {b.tag:UOp.placeholder((sizes[b.tag],), b.dtype, next(UOp.unique_num), device=b.device).rtag(b.tag) for b in bufs if counts[b.tag] > 1}
-  subs = {b:UOp(Ops.SLICE, b.dtype, (bases[b.tag], UOp.const(offs.get(b, 0))), b.max_numel()) for b in bufs if b.tag in bases}
+  subs = {b:bases[b.tag][(off:=offs.get(b, 0)):off+b.max_numel()] for b in bufs if b.tag in bases}
   return call.replace(src=(call.src[0].substitute(subs, walk=True), *call.src[1:])) if subs else None
 pm_pack_placeholders = PatternMatcher([
   (UPat(Ops.CALL, src=(UPat(Ops.CUSTOM_FUNCTION, arg="hcq"),), name="call", allow_any_len=True), pack_hcq_placeholders)])
@@ -481,7 +494,7 @@ pm_resolve_patches = PatternMatcher([
   (UPat(name="buf").index(UPat(Ops.RANGE), allow_any_len=True)
     .store(UPat.any(UPat(Ops.BINARY, name="blob"), UPat(Ops.BINARY, name="blob").bitcast()).index(UPat(Ops.RANGE), allow_any_len=True).load())
     .end(UPat(Ops.RANGE)), fold_binary),
-  (UPat({Ops.BUFFER, Ops.SLICE, Ops.MSTACK}, name="buf").index(UPat(Ops.STACK, name="off")).store(UPat(Ops.STACK, name="val")), fold_const_store),
+  (UPat({Ops.BUFFER, Ops.MSTACK}, name="buf").index(UPat(Ops.STACK, name="off")).store(UPat(Ops.STACK, name="val")), fold_const_store),
 ])
 
 pm_assert_no_afters = PatternMatcher([(UPat(Ops.AFTER, name="a"), lambda a: panic(RuntimeError, f"AFTER left at hcq_link: {a.src[0].op}"))])
@@ -611,6 +624,8 @@ class HCQAllocator(LRUAllocator[HCQDeviceType], Generic[HCQDeviceType]):
     if not hasattr(self, '_do_map'): raise NotImplementedError("map failed: no method implemented")
     return self._do_map(buf)
 
+  def _do_unmap(self, mb): self.dev.iface.free(mb)
+
   @suppress_finalizing
   def _free(self, buf:HCQ2Buffer, options:BufferSpec|None=None):
     if options is not None and options.external_ptr is not None: return
@@ -619,6 +634,6 @@ class HCQAllocator(LRUAllocator[HCQDeviceType], Generic[HCQDeviceType]):
 
   def _unmap(self, mb):
     self.dev.synchronize()
-    self.dev.iface.free(mb)
+    self._do_unmap(mb)
 
   def _offset(self, buf, size:int, offset:int) -> HCQ2Buffer: return buf.offset(offset=offset, size=size)

tinygrad/schedule/__init__.py

@@ -81,7 +81,8 @@ def create_schedule(sched_sink:UOp) -> UOp:
 
 from tinygrad.schedule.memory import memory_plan_rewrite
 from tinygrad.engine.realize import capturing, pm_flatten_linear
-from tinygrad.schedule.rangeify import get_kernel_graph
+#from tinygrad.schedule.rangeify import get_kernel_graph
+from tinygrad.schedule.rangeify2 import get_kernel_graph
 from tinygrad.helpers import CAPTURING
 from tinygrad.uop.ops import PatternMatcher, UPat, ParamArg
 from tinygrad.dtype import AddrSpace

tinygrad/schedule/indexing.py

@@ -23,7 +23,7 @@ class IndexingContext:
     return UOp.range(s, next(self.range_idx), axistype) if resolve(s!=1) else UOp.const(0)
 
 
-ALWAYS_CONTIGUOUS: set[Ops] = {Ops.CONTIGUOUS, Ops.AFTER, Ops.BUFFER, Ops.SLICE,
+ALWAYS_CONTIGUOUS: set[Ops] = {Ops.CONTIGUOUS, Ops.AFTER, Ops.BUFFER,
                       Ops.CONST, Ops.BIND, Ops.MSELECT, Ops.MSTACK, Ops.PARAM,
                       Ops.LOAD, Ops.CALL, Ops.FUNCTION}
 
@@ -34,10 +34,6 @@ def realize_srcs(ctx:IndexingContext, rb:UOp) -> None:
     if s.base.op not in ALWAYS_CONTIGUOUS: ctx.realize_map[s] = None
 
 def realize_store_after_src(ctx:IndexingContext, dest:UOp, src:UOp):
-  # don't realize SLICE when it's the direct source of STORE+AFTER — the target buffer is the output
-  if src.op is Ops.SLICE and src in ctx.realize_map \
-     and not dest.op_in_backward_slice_with_self(Ops.SHRINK, Ops.PERMUTE, Ops.FLIP, Ops.PAD):
-    del ctx.realize_map[src]
   # you don't usually have to do this for assign unless there's a WAR hazard like TestAssign.test_assign_double_diamond_reduce
   if dest.base in src.backward_slice_with_self: ctx.realize_map[src] = None
 
@@ -74,7 +70,7 @@ def broadcast_rngs(x:UOp, src:UOp, rngs:tuple[UOp, ...]) -> tuple[UOp, ...]:
 # TODO: srcs contain (real data srcs, something else, ranges) and the boundary is confusing. see range_start
 def data_srcs(op:Ops, src:tuple[UOp, ...]) -> tuple[UOp, ...]:
   if op in {Ops.PARAM, Ops.BUFFER, Ops.RANGE, Ops.SPECIAL, Ops.BIND}: return ()
-  if op in GroupOp.Movement|{Ops.INDEX, Ops.SLICE, Ops.STAGE, Ops.REDUCE, Ops.AFTER, Ops.END}: return src[:1]
+  if op in GroupOp.Movement|{Ops.INDEX, Ops.STAGE, Ops.REDUCE, Ops.AFTER, Ops.END}: return src[:1]
   return src
 
 def create_bufferize_and_index_srcs(ctx:IndexingContext, x:UOp) -> list[UOp]:
@@ -84,7 +80,7 @@ def create_bufferize_and_index_srcs(ctx:IndexingContext, x:UOp) -> list[UOp]:
   for i, s in enumerate(x.src):
     new_src = s
     src_rngs = broadcast_rngs(x, s, ctx.range_map[x][0]) if x in ctx.range_map else ()
-    if s.op in {Ops.PARAM, Ops.BUFFER, Ops.SLICE, Ops.MSTACK, Ops.MSELECT, Ops.AFTER}:
+    if s.op in {Ops.PARAM, Ops.BUFFER, Ops.MSTACK, Ops.MSELECT, Ops.AFTER}:
       if x in ctx.range_map and i < data_src_count: new_src = new_src.index(*src_rngs)
     elif s in ctx.realize_map:
       realized_ranges = ctx.realize_map[s]

tinygrad/schedule/memory.py

@@ -52,11 +52,9 @@ def memory_plan_rewrite(linear:UOp, held_bufs:set[UOp]|None=None) -> UOp:
     peaks[_key(buf)] = (max(peaks[_key(buf)][0], offsets[buf] + buf.max_numel() * buf.dtype.itemsize), peaks[_key(buf)][1])
   arena_sizes = {key: round_up(peak, block_size) for key, (peak, _) in peaks.items()}
 
-  # build replace_map: each buffer becomes a SLICE into a shared per-device-lane arena
+  # build replace_map: each buffer becomes a SHRINK/BITCAST into a shared per-device-lane arena
   arenas = {key: UOp.new_buffer(key[0], sz, dtypes.int8) for key, sz in arena_sizes.items()}
-  replace_map:dict[UOp, UOp] = {}
-  for buf_uop, offset in offsets.items():
-    replace_map[buf_uop] = UOp(Ops.SLICE, buf_uop.dtype, (arenas[_key(buf_uop)], UOp.const(offset)), buf_uop.max_numel())
+  replace_map = {buf_uop:arenas[_key(buf_uop)][offset:offset+buf_uop.nbytes()].bitcast(buf_uop.dtype) for buf_uop, offset in offsets.items()}
 
   if DEBUG >= 1 and (omem:=sum(nbytes.values()) / 1e6) != (nmem:=sum(arena_sizes.values()) / 1e6):
     print(f"memory reduced from {omem:.2f} MB -> {nmem:.2f} MB, {len(first_appearance)} -> {len(arenas)} bufs")

tinygrad/schedule/rangeify2.py

@@ -0,0 +1,238 @@
+from dataclasses import dataclass, field, replace
+from typing import cast
+import itertools
+from tinygrad.dtype import dtypes, AddrSpace, Invalid, to_dtype, strong_dtype
+from tinygrad.uop.ops import PatternMatcher, UPat, Ops, UOp, resolve, GroupOp, KernelInfo, ParamArg, shape_to_shape_arg
+from tinygrad.uop.ops import graph_rewrite, sint, AxisType, BottomUpGate, rewrite_group, identity_element, remove_all_tags
+from tinygrad.uop.symbolic import symbolic, pm_fold_cast_const
+from tinygrad.uop.movement import mop_cleanup
+from tinygrad.helpers import prod, getenv, dedup, all_int, DEBUG, SPLIT_REDUCEOP, DEBUG_RANGEIFY, VIZ, MAX_KERNEL_BUFFERS, SPEC
+from tinygrad.helpers import PCONTIG, FLOAT16, OPENPILOT_HACKS, argsort, partition, get_single_element, Context
+from tinygrad.codegen.simplify import pm_flatten_range, pm_reduce_simplify
+from tinygrad.codegen.opt import Opt
+from tinygrad.schedule.indexing import run_rangeify, BufferizeOpts, IndexingContext, apply_movement_op
+from tinygrad.schedule.multi import multi_pm
+from tinygrad.schedule.allreduce import create_allreduce_function
+
+# *** preparation ***
+
+from tinygrad.helpers import all_same
+from tinygrad.uop.ops import _broadcast_shape
+
+def expand_broadcast(x:UOp):
+  shapes = [u._shape for u in x.src]
+  if any(s is None for s in shapes) or all_same(shapes): return None
+  shape = _broadcast_shape(*shapes)
+  return x.replace(src=tuple([u.expand(shape) for u in x.src]))
+
+pm_expand_broadcast = PatternMatcher([
+  # expand broadcasts first
+  (UPat(GroupOp.Binary|GroupOp.Ternary|{Ops.STORE}, name="x"), expand_broadcast),
+])
+
+def convert_copy_to_store(ctx, copy:UOp, existing_buf:UOp|None=None):
+  input_src = copy.src[0]
+  if not input_src.has_buffer_identity(after_ok=True): input_src = input_src.contiguous()
+  input_src = input_src.flatten()
+  if existing_buf is not None:
+    # if the existing buffer is not a full buffer, we can't use it
+    if not existing_buf.has_buffer_identity(after_ok=True): return None
+    # if there's already a buffer, we just use it
+    return existing_buf.flatten().store(input_src)
+  # create the output buffer
+  buf = UOp(Ops.BUFFER, src=(shape_to_shape_arg(input_src.max_shape),), arg=ParamArg(next(ctx), copy.dtype, device=copy.device))
+  # reshape back to input
+  return buf.after(buf.store(input_src)).reshape(copy.shape)
+
+def convert_contig_to_store(ctx, copy:UOp):
+  input_src = copy.src[0]
+  # create the output buffer
+  buf = UOp(Ops.BUFFER, src=(shape_to_shape_arg(input_src.max_shape),), arg=ParamArg(next(ctx), copy.dtype, device=copy.device))
+  # reshape back to input
+  view = buf.shrink_to(input_src.shape)
+  return view.after(view.store(input_src))
+
+pm_copy_to_store = PatternMatcher([
+  (UPat(name="existing_buf").store(UPat(Ops.COPY, name="copy")), convert_copy_to_store),
+  (UPat(Ops.COPY, name="copy"), convert_copy_to_store),
+  (UPat(Ops.CONTIGUOUS, name="copy"), convert_contig_to_store),
+])
+
+# *** RANGE creation ***
+
+def rangeify_on_reduce(ctx, inp:UOp, red:UOp, idx:UOp|None=None):
+  if red.arg[1] == 0: return None
+  if idx is None and len(red.shape) > 0: return None
+  # TODO: is AxisType.REDUCE a real thing?
+  rngs = [UOp.range(s, next(ctx), AxisType.REDUCE) for s in inp.shape[:red.arg[1]]]
+  return inp.index(*rngs, *(idx.src[1:] if idx is not None else ())).reduce(*rngs, arg=(red.arg[0], 0))
+
+def rangeify_on_store(ctx, x:UOp):
+  if x.shape == (): return None
+  rngs = [UOp.range(s, next(ctx)) for s in x.shape]
+  return x.src[0].index(*rngs).store(x.src[1].index(*rngs)).end(*rngs)
+
+def rangeify_on_stage(ctx, x:UOp):
+  if x.src[0].shape == (): return None
+  # size 1 dims don't get ranges, they are reshaped out and back in
+  if all_int(x.shape) and 0 < len(sq := tuple(s for s in x.shape if s != 1)) < len(x.shape):
+    return rangeify_on_stage(ctx, x.src[0].reshape(sq).bufferize(arg=x.arg)).reshape(x.shape)
+  rngs = [UOp.range(s, next(ctx)) for s in x.shape]
+  return x.replace(src=(x.src[0].index(*rngs), *rngs))
+
+pm_range_creation = PatternMatcher([
+  # reduce/store are what creates ranges
+  (UPat(Ops.REDUCE, src=(UPat.var('inp'),), name="red").index(name="idx", allow_any_len=True), rangeify_on_reduce),
+  (UPat(Ops.REDUCE, src=(UPat.var('inp'),), name="red"), rangeify_on_reduce),
+  (UPat(Ops.STORE, name="x"), rangeify_on_store),
+  (UPat(Ops.STAGE, name="x"), rangeify_on_stage),
+])
+
+# *** RANGE migration ***
+
+# movement op on INDEX as a PatternMatcher
+def _mop_index(r:UOp, idx:UOp):
+  idxs = idx.src[1:]
+  if len(idxs) == len(r.shape):
+    ret = r.src[0].index(*apply_movement_op(r.op, r.src[0].shape, r.marg, idxs), dtype=idx.dtype, arg=idx.arg)
+    if r.op is Ops.PAD:
+      # insert 0 for PAD with where
+      # TODO: does this need simplify to ensure the Invalids are at the base?
+      a = UOp.const(True)
+      for s in ret.src[1:]:
+        if s.op is Ops.WHERE and s.src[2].op is Ops.CONST and s.src[2].arg == Invalid: a = a & s.src[0]
+      ret = a.where(ret, ret.const_like(0))
+    return ret
+  if r.op is Ops.RESHAPE:
+    src_prefix = len(r.src[0].shape) - len(r.shape[len(idxs):])
+    if src_prefix >= 0 and r.src[0].shape[src_prefix:] == r.shape[len(idxs):]:
+      if src_prefix == 0: return r.src[0] if r.src[0].dtype == idx.dtype else None
+      ret = r.src[0].index(*apply_movement_op(r.op, r.src[0].shape[:src_prefix], r.shape[:len(idxs)], idxs), dtype=idx.dtype, arg=idx.arg)
+      return ret if ret.shape == idx.shape else None
+
+# TODO: this should be in _mop_index
+def index_on_stack(stack:UOp, idx:UOp):
+  srcs = [s.index(*idx.src[2:]) for s in stack.src]
+  r0 = idx.src[1]
+  ret = srcs[-1]
+  for k in range(len(srcs)-2, -1, -1): ret = r0.eq(k).where(srcs[k], ret)
+  return ret
+
+def walk_mop(u:UOp):
+  if u.op in GroupOp.Movement or u.op is Ops.INDEX: return u.src[0]
+  assert u.op == Ops.AFTER
+  return u
+
+pm_range_migration = PatternMatcher([
+  # INDEX without src is nothing
+  (UPat(Ops.INDEX, src=(UPat.var('x'),)), lambda x: x),
+  # STAGE on shape () is nothing
+  (UPat(Ops.STAGE, src=(UPat.var('x'),)), lambda x: x if x.shape == () else None),
+  # if INDEX is on STAGE with the same ranges, remove the pair
+  (UPat(Ops.STAGE, allow_any_len=True, name="s").index(allow_any_len=True, name="i"),
+   lambda s,i: s.src[0] if s.src[1:] == i.src[1:] else None),
+  # reshape of a single element shaped value to scalar is an index
+  (UPat(Ops.RESHAPE, name="x"), lambda x: x.src[0].index(0) if x.marg == () and x.src[0].shape == (1,) else None),
+  # handle movement ops on INDEX
+  (UPat(GroupOp.Movement, name="r").index(name="idx", allow_any_len=True), _mop_index),
+  (UPat(Ops.STACK, name="stack").index(name="idx", allow_any_len=True), index_on_stack),
+  # move movement ops and INDEX after AFTER
+  (UPat(GroupOp.Movement|{Ops.INDEX}, name="r").after(name="a", allow_any_len=True),
+   lambda r,a: UOp(r.op, src=(a.replace(src=(r.src[0],)+a.src[1:]),)+r.src[1:], arg=r.arg)),
+  # pass index through elementwise
+  (UPat(GroupOp.Elementwise, name="b").index(name="idx", allow_any_len=True),
+   lambda b,idx: b.replace(src=tuple(s.index(*idx.src[1:]) for s in b.src))),
+  # remove movement ops from SINK. TODO: should be generic
+  (UPat(Ops.SINK, name="s"), lambda s: s.replace(src=tuple(walk_mop(u) for u in s.src))),
+])
+
+# *** split into kernels ***
+
+@dataclass
+class SplitCtx:
+  call_args:list = field(default_factory=list)
+  range_number:int = -1
+
+def _split_graph(ctx:SplitCtx, u:UOp) -> UOp:
+  assert len(u.shape) <= 1, f"rangeify needs to reduce to a single idx, not {u.shape}"
+  ctx.call_args.append(u)
+  return u.param_like(len(ctx.call_args)-1)
+
+def _renumber_range(ctx:SplitCtx, u:UOp) -> UOp:
+  ctx.range_number += 1
+  return u.replace(arg=(ctx.range_number, u.arg[-1]))
+
+pm_split_graph = PatternMatcher([
+  (UPat((Ops.PARAM, Ops.AFTER, Ops.BUFFER), name="u"), _split_graph),
+  (UPat(Ops.RANGE, name="u"), _renumber_range),
+])
+
+def split_store(x:UOp) -> UOp:
+  ret = graph_rewrite(x, pm_split_graph, ctx:=SplitCtx(), name="split kernel", bottom_up=True, walk=True)
+  return ret.sink(arg=KernelInfo()).call(*ctx.call_args)
+
+split_kernels = PatternMatcher([
+  (UPat((Ops.STORE, Ops.END), name="x"), split_store),
+])
+
+# *** main rangeify ***
+
+debug_tag_factor = PatternMatcher([
+  (UPat(GroupOp.All, name="x"), lambda ctx,x: x.rtag(ctx[0][x] if x not in ctx[1] else 'REAL') if x.tag is None else None),
+])
+
+def remove_stage(ctx, x:UOp) -> UOp:
+  buf = UOp.new_buffer(x.arg.device, x.max_numel(), x.dtype, num=next(ctx))
+  return buf.after(buf.reshape(x.shape).index(*x.src[1:]).store(x.src[0]).end(*x.src[1:])).reshape(x.shape)
+pm_remove_stage = PatternMatcher([(UPat(Ops.STAGE, name="x"), remove_stage)])
+
+@rewrite_group(new_ctx=False)
+def get_kernel_graph(sink:UOp) -> UOp:
+  # TODO: multi should just be part of rangeify
+  tsink = graph_rewrite(sink, multi_pm, name="multi_pm")
+
+  # prepare
+  tsink = graph_rewrite(tsink, pm_expand_broadcast, bottom_up=True, name="expand broadcast")
+  tsink = graph_rewrite(tsink, pm_copy_to_store, ctx=itertools.count(0), bottom_up=True, name="convert copy to store")
+
+  # add safe STAGEs to never duplicate compute
+  # we compute the number of times a buffer is consumed. if > 1, we realize
+  realize = {}
+  consumes = {tsink:0}
+  for u in reversed(tsink.toposort()):
+    assert u in consumes, f"{u.op} not in consumes"
+    if (u.op in GroupOp.ALU or u.op is Ops.REDUCE) and consumes[u] > 1 and u.device is not None:
+      # TODO: rename to stage
+      realize[u] = u.rtag(1).bufferize(arg=BufferizeOpts(device=u.device))
+      consumes[u] = 1
+    if u.op is Ops.STORE: consumes[u] = 1
+    if u.op is Ops.EXPAND: consumes[u] *= u.max_numel() // u.src[0].max_numel()
+    for i,s in enumerate(u.src):
+      if s not in consumes: consumes[s] = 0
+      if u.op is not Ops.STORE or i > 0:
+        consumes[s] += consumes[u]
+  if VIZ:
+    with Context(TRACK_MATCH_STATS=0): ctags = graph_rewrite(tsink, debug_tag_factor, ctx=(consumes, realize), bottom_up=True)
+    graph_rewrite(ctags, PatternMatcher([]), name="View Consumes")
+
+  # add stages
+  tsink = graph_rewrite(tsink.substitute(realize), remove_all_tags, name="untag")
+
+  # simple rangeify
+  tsink = graph_rewrite(tsink, pm_range_creation+pm_range_migration, ctx=itertools.count(0), bottom_up=True, name="simple rangeify")
+
+  # TODO: merging and splitting algorithm
+
+  if VIZ: graph_rewrite(tsink, PatternMatcher([]), name="View Rangeify")
+
+  tsink = graph_rewrite(tsink, pm_remove_stage, ctx=itertools.count(0), bottom_up=True, name="remove stage")
+
+  tsink = graph_rewrite(tsink, split_kernels, bottom_up=True, name="split kernels")
+
+  if VIZ: graph_rewrite(tsink, PatternMatcher([]), name="View Kernel Graph")
+  if SPEC:
+    # validate the kernel graph
+    from tinygrad.uop.spec import type_verify, spec_kernel_graph
+    type_verify(tsink, spec_kernel_graph, enter_calls=False)
+  return tsink
+

tinygrad/tensor.py

@@ -23,6 +23,7 @@ class AllocCtx:
   bases: set[UOp] = field(default_factory=set)
   assigns: list[UOp] = field(default_factory=list)
   replacements: list[UOp] = field(default_factory=list)
+  views: set[UOp] = field(default_factory=set)
 
 def tag_uop(ctx:AllocCtx, x:UOp):
   if x.tag is not None: return None
@@ -63,40 +64,37 @@ def replace_contig_with_store_after(u:UOp):
 def replace_store_after_with_contig(u:UOp, src:UOp):
   assigned_to = u
   while assigned_to.op in {Ops.BITCAST, Ops.AFTER, Ops.UNSHARD}: assigned_to = assigned_to.src[0].base
-  if assigned_to.op not in {Ops.BUFFER, Ops.SLICE}: return src.contiguous(tag=u.tag)
+  if assigned_to.op is not Ops.BUFFER: return src.contiguous(tag=u.tag)
 
 def _make_buffer_view(src:UOp) -> UOp|None:
-  """If movement ops on src collapse to a contiguous range, return SLICE. Otherwise None."""
-  if (offset := src.contiguous_view_offset()) is None: return None
-  buf = src.base
-  if buf.op is Ops.SLICE:
-    byte_offset = buf.src[1].val * buf.src[0].dtype.itemsize + offset * src.dtype.itemsize
-    buf = buf.src[0]
-    if byte_offset % buf.dtype.itemsize != 0: return None
-    offset = byte_offset // buf.dtype.itemsize
-  return UOp(Ops.SLICE, src.dtype, (buf, UOp.const(offset)), src.numel())
+  if (cv := src.contiguous_view()) is None: return None
+  (buf, offset), size = cv, src.max_numel() * src.element_size() // cv[0].element_size()
+  if buf.op is not Ops.BUFFER: return None
+  # NB: make offset a UOp.variable here to do the offset computation in the kernels
+  return buf[offset:offset+size].bitcast(src.dtype)
 
-def contiguous_mops_to_view(c:UOp, src:UOp):
-  """MOPS(BUFFER) → SLICE when movement ops collapse to a contiguous range."""
+def contiguous_mops_to_view(ctx:AllocCtx, c:UOp, src:UOp):
+  """MOPS(BUFFER) → SHRINK when movement ops collapse to a contiguous range."""
   buf = src.base
-  if buf.op not in {Ops.BUFFER, Ops.SLICE, Ops.UNSHARD}: return None
-  if src.op is Ops.RESHAPE and src.src[0].op in {Ops.BUFFER, Ops.SLICE} and c.op is not Ops.BITCAST: return None
-  if c.op is not Ops.BITCAST and src.op is Ops.BUFFER: return None
+  while buf.op is Ops.BITCAST: buf = buf.src[0].base
+  if buf.op not in {Ops.BUFFER, Ops.UNSHARD}: return None
 
   # no symbolic shape
   if not all_int(c.shape): return None
 
   if buf.op is not Ops.UNSHARD and (view := _make_buffer_view(src)) is not None:
-    view = (view.replace(dtype=c.dtype, arg=c.numel()) if c.op is Ops.BITCAST else view).reshape(c.shape)
-    return c.replace(src=(view,)) if c.op is Ops.COPY else view
+    ctx.views.add(view)
+    view = view.reshape(c.shape)
+    return c.replace(src=(view,)+c.src[1:]) if c.op in {Ops.COPY, Ops.STORE} else view
 
-  # for UNSHARD tensors, use multi_pm to resolve per-shard movement ops, then create SLICE on the resolved result
+  # for UNSHARD tensors, use multi_pm to resolve per-shard movement ops, then create SHRINK on the resolved result
   if not isinstance(c.device, str):
     from tinygrad.schedule.multi import multi_pm
     resolved = graph_rewrite(src, multi_pm, name="multi_buffer_view")
     if resolved.op is not Ops.UNSHARD: return None
     if (view := _make_buffer_view(resolved.src[0])) is None: return None
-    return view.reshape(resolved.src[0].shape).unshard(resolved.arg, resolved.src[1:]).contiguous(tag=c.tag)
+    ctx.views.add(view)
+    return view.reshape(resolved.src[0].shape).unshard(resolved.arg, resolved.src[1:])
 
   return None
 
@@ -151,8 +149,9 @@ pm_early_transform_tensor_graph = PatternMatcher([
   # resolve TUPLE+GETTUPLE (for precompiled calls)
   (UPat(Ops.GETTUPLE, src=(UPat(Ops.TUPLE, name="t"),), name="g"), lambda g,t: t.src[g.arg]),
 
-  # fold MOPS+BITCAST over BUFFER/SLICE into SLICE when movement ops collapse to contiguous range
-  (UPat((Ops.BITCAST, Ops.COPY, Ops.CONTIGUOUS), src=(UPat(GroupOp.Movement|{Ops.BUFFER}, name="src"),), name="c"), contiguous_mops_to_view),
+  # fold MOPS+BITCAST over BUFFER into SHRINK when movement ops collapse to contiguous range
+  (UPat((Ops.COPY, Ops.CONTIGUOUS), src=(UPat(GroupOp.Movement|{Ops.BITCAST}, name="src"),), name="c"), contiguous_mops_to_view),
+  (UPat(Ops.STORE, src=(UPat(Ops.BITCAST, name="src"), UPat()), name="c", allow_any_len=True), contiguous_mops_to_view),
 
   # remove contiguous on movement ops before a copy on disk
   (UPat(GroupOp.Movement-{Ops.SHRINK, Ops.RESHAPE}, name="x").f(Ops.CONTIGUOUS).f(Ops.COPY, name="copy"), lambda x,copy:
@@ -201,6 +200,8 @@ def replace_input_buffer(ctx:AllocCtx, b:UOp):
   return UOp.param(len(ctx.replacements)-1, b.dtype, b.shape, b.device,
                    addrspace=b.addrspace if b.addrspace is not None else AddrSpace.GLOBAL)
 
+def replace_input_view(ctx:AllocCtx, b:UOp): return replace_input_buffer(ctx, b) if b in ctx.views else None
+
 pm_finalize_call = PatternMatcher([
   (UPat(Ops.AFTER, name="x"), finalize_after),
   (UPat(Ops.COPY, name="x"), lambda ctx,x: ctx.assigns.append(x) if isinstance(x.device, str) and x.device.startswith(("DISK", "TINYFS")) else None),
@@ -210,8 +211,9 @@ pm_replace_buf = PatternMatcher([
   # replace BUFFER with PARAM for cache key normalization
   (UPat(Ops.BUFFER, src=(UPat(),), name="b"), lambda ctx,b:
    replace_input_buffer(ctx, b) if isinstance(b.arg, ParamArg) and b.addrspace is AddrSpace.GLOBAL else None),
-  # replace SLICE with PARAM. this rewrite is bottom up so BUFFERs we don't need won't be in the input
-  (UPat(Ops.SLICE, src=(UPat(Ops.BUFFER), UPat(Ops.CONST, dtype=dtypes.weakint)), name="b"), replace_input_buffer),
+  # replace SHRINK with PARAM
+  (UPat(Ops.SHRINK, src=(UPat(Ops.BUFFER),), name="b", allow_any_len=True), replace_input_view),
+  (UPat(Ops.BITCAST, src=(UPat.any(UPat(Ops.SHRINK, src=(UPat(Ops.BUFFER),), allow_any_len=True), UPat(Ops.BUFFER)),), name="b"), replace_input_view),
   # strip value from BIND for cache key normalization, so different values hit same cache
   (UPat(Ops.BIND, src=(UPat(Ops.PARAM), UPat(Ops.CONST)), name="b"), replace_input_buffer),
 ])
@@ -229,7 +231,7 @@ def transform_to_call(big_sink:UOp) -> tuple[UOp, dict[UOp, UOp]]:
   big_sink = graph_rewrite(big_sink, add_tags, ctx=ctx, bottom_up=True, name="number the uops")
 
   # here we can break the tensor graph. this is the only place you need to maintain numbered tags
-  big_sink = graph_rewrite(big_sink, pm_early_transform_tensor_graph, name="early transform tensor graph")
+  big_sink = graph_rewrite(big_sink, pm_early_transform_tensor_graph, ctx=ctx, name="early transform tensor graph")
 
   # here we construct the final buffer_map: as-built nodes -> their final storage. values are never keys
   graph_rewrite(big_sink, pm_finalize_call, ctx=ctx, name="finalize call")

tinygrad/uop/__init__.py

@@ -93,7 +93,7 @@ class Ops(FastEnum):
   CONTIGUOUS = auto(); CONTIGUOUS_BACKWARD = auto(); DETACH = auto()
 
   # buffer ops
-  STAGE = auto(); COPY = auto(); SLICE = auto(); MSELECT = auto(); MSTACK = auto(); CUSTOM_FUNCTION = auto()
+  STAGE = auto(); COPY = auto(); MSELECT = auto(); MSTACK = auto(); CUSTOM_FUNCTION = auto()
 
   # the core 6 movement ops! these only exist in the tensor graph
   RESHAPE = auto(); PERMUTE = auto(); EXPAND = auto(); PAD = auto(); FLIP = auto()

tinygrad/uop/ops.py

@@ -45,8 +45,7 @@ axis_colors = {AxisType.DEVICE: "green", AxisType.GLOBAL: "blue", AxisType.THREA
 axis_to_pos = {AxisType.DEVICE: -2, AxisType.WEAK: -1, AxisType.LOOP: -1, AxisType.THREAD: 0, AxisType.GLOBAL: 0, AxisType.WARP: 1,
                AxisType.LOCAL: 2, AxisType.UPCAST: 3, AxisType.GROUP_REDUCE: 2, AxisType.REDUCE: 4, AxisType.UNROLL: 5}
 
-range_start = {Ops.STAGE: 1, Ops.REDUCE: 1, Ops.WMMA: 3, Ops.END: 1, Ops.CALL: 1, Ops.FUNCTION: 1,
-               Ops.SLICE: 2, Ops.LINEAR: 0}
+range_start = {Ops.STAGE: 1, Ops.REDUCE: 1, Ops.WMMA: 3, Ops.END: 1, Ops.CALL: 1, Ops.FUNCTION: 1, Ops.LINEAR: 0}
 
 # https://en.wikipedia.org/wiki/Identity_element
 def identity_element(op:Ops, dt:DType) -> PyConst: return dt.const({Ops.ADD:0, Ops.MUL:1, Ops.MAX:dt.min}[op])
@@ -171,9 +170,6 @@ def dtype_from_uop(op:Ops, src:tuple[UOp,...], arg:Any) -> DType|None:
       return arg.dtype
     case Ops.BINARY:
       return dtypes.uint8
-    case Ops.SLICE:
-      # TODO: slice just shouldn't exist
-      return None
     case Ops.CAST | Ops.BITCAST:
       assert isinstance(arg, DType), f"CAST/BITCAST arg must be DType, got {arg}"
       return arg
@@ -221,7 +217,7 @@ class UOpMetaClass(type):
     return created
 
 # some uops map to other stuff
-buffers:weakref.WeakKeyDictionary[UOp, Buffer|MultiBuffer] = weakref.WeakKeyDictionary() # this maps BUFFER/SLICE uops to their device Buffers
+buffers:weakref.WeakKeyDictionary[UOp, Buffer|MultiBuffer] = weakref.WeakKeyDictionary() # this maps BUFFER/view uops to their device Buffers
 all_metadata:weakref.WeakKeyDictionary[UOp, tuple[Metadata, ...]] = weakref.WeakKeyDictionary() # TODO: should this be here?
 
 # recursive_property replaces functools.cached_property in recursive UOp functions to prevent RecursionError
@@ -386,10 +382,6 @@ class UOp(RandMixin, metaclass=UOpMetaClass):
       case Ops.BUFFER:
         if len(self.src): return self.src[0].as_shape
         return ()
-      case Ops.SLICE:
-        # HACK: SLICE is used inside kernels, so we set the shape to () if it's on an INDEX
-        if self.src[0].op is Ops.INDEX: return ()
-        return (self.arg,)
       case Ops.CUSTOM | Ops.CUSTOMI:
         if self.dtype is dtypes.void: return None
         input_shapes = [x._shape for x in self.src if x._shape is not None]
@@ -568,9 +560,9 @@ class UOp(RandMixin, metaclass=UOpMetaClass):
     in_tuple = self.src[0] if self.op is Ops.FUNCTION else self
     assert in_tuple.op is Ops.TUPLE, f"gettuple requires FUNCTION or TUPLE source, got {self.op}"
     return UOp(Ops.GETTUPLE, src=(self,), arg=idx)
-  def group(*srcs:UOp|None):  # pylint: disable=no-self-argument
+  def group(*srcs:UOp|None, **kwargs):  # pylint: disable=no-self-argument
     if len(srcs) == 1 and isinstance(srcs[0], UOp): return srcs[0]
-    return UOp(Ops.GROUP, src=tuple([x for x in srcs if x is not None]))
+    return UOp(Ops.GROUP, src=tuple([x for x in srcs if x is not None]), **kwargs)
   def index(self, *srcs:UOp|int|None, **kwargs):
     new_srcs: list[UOp] = [UOp.const(x) if isinstance(x, int) else x for x in srcs if x is not None]
     if len(new_srcs) == 1 and new_srcs[0].op is Ops.CONST and self.op is Ops.STACK: return self.src[new_srcs[0].val]
@@ -822,7 +814,7 @@ class UOp(RandMixin, metaclass=UOpMetaClass):
   unique_num = itertools.count(0)
 
   def getaddr(self, device=None) -> UOp:
-    if self.without_after.op not in {Ops.BUFFER, Ops.SLICE, Ops.BINARY, Ops.MSTACK, Ops.MSELECT, Ops.PARAM}: return self
+    if self.without_after.op not in {Ops.BUFFER, Ops.SHRINK, Ops.BITCAST, Ops.BINARY, Ops.MSTACK, Ops.MSELECT, Ops.PARAM}: return self
     return UOp(Ops.GETADDR, src=(self,), arg=device or to_tuple(self.device)[0])
   @staticmethod
   def new_buffer(device:str|tuple[str, ...], size:int, dtype:DType, num=None):
@@ -901,8 +893,7 @@ class UOp(RandMixin, metaclass=UOpMetaClass):
     while len(s.src) and s.op not in {Ops.BUFFER, Ops.PARAM, Ops.STAGE, Ops.MSTACK}: s = s.src[0]
     return s
 
-  def contiguous_view_offset(self) -> int|None:
-    """If movement ops on a BUFFER collapse to a contiguous range, return `offset` in elements. Otherwise None."""
+  def contiguous_view(self) -> tuple[UOp, int]|None:
     from tinygrad.schedule.rangeify import pm_mops
     from tinygrad.uop.symbolic import symbolic
 
@@ -915,14 +906,17 @@ class UOp(RandMixin, metaclass=UOpMetaClass):
 
     idx = self.flatten().index(UOp.range(self.numel(), 0))
     out = graph_rewrite(idx, pm_mops+symbolic+pm_contiguous_view_offset, ctx=self, name="contiguous_view_offset")
-    return out.val if out.op is Ops.CONST and isinstance(out.val, int) else None
+    if out.op is not Ops.INDEX or not (b:=out.src[0]).tag or (c:=out.src[1]).op is not Ops.CONST or not isinstance(c.val, int): return None
+    return b.rtag(None), c.val
+
+  def contiguous_view_offset(self) -> int|None: return None if (view := self.contiguous_view()) is None else view[1]
 
   def has_buffer_identity(self, after_ok=False):
     """Check if this UOp has a concrete buffer identity in the graph (RESHAPE/UNSHARD -> BUFFER chain)."""
     # TODO: this is confusing because UOp.variable('v', 0, 1, dtypes.weakfloat) is True for jit to work, but it doesn't have a buffer
     if self.op in {Ops.RESHAPE, Ops.UNSHARD, Ops.MSELECT}: return self.src[0].has_buffer_identity(after_ok)
     if after_ok and self.op == Ops.AFTER: return self.src[0].has_buffer_identity(after_ok)
-    return self.op in {Ops.BUFFER, Ops.SLICE, Ops.PARAM}
+    return self.op in {Ops.BUFFER, Ops.PARAM}
 
   def _base_buffer_is_realized(self) -> bool:
     """Walk through AFTER chain to find if the underlying buffer is realized (has allocated memory)."""
@@ -932,30 +926,19 @@ class UOp(RandMixin, metaclass=UOpMetaClass):
 
   @property
   def buffer(self) -> Buffer|MultiBuffer:
-    if self.op in {Ops.CONTIGUOUS, Ops.RESHAPE, Ops.UNSHARD, Ops.DETACH, Ops.AFTER}: return self.src[0].buffer
+    if self.op in {Ops.CONTIGUOUS, Ops.CONTIGUOUS_BACKWARD, Ops.RESHAPE, Ops.UNSHARD, Ops.DETACH, Ops.AFTER}: return self.src[0].buffer
     # this buffer can process disk tensors and simple movement ops
-    if self is not self.base:
-      buf = self.base.buffer
-      assert isinstance(buf, Buffer), "must be a Buffer for movement ops"
-      offset = self.contiguous_view_offset()
-      if offset is None: raise RuntimeError(f"non-contiguous view is not supported for {buf.device} buffer")
-      return buf.view(prod(self.max_shape), self.dtype, offset*self.dtype.itemsize)
-    if self.op is Ops.BITCAST:
-      buf = self.src[0].buffer
-      assert isinstance(buf, Buffer), "must be a Buffer for BITCAST"
-      return buf.view(prod(self.max_shape), self.dtype, 0)
-    if self.op is Ops.SLICE:
+    if self is not self.base or self.op is Ops.BITCAST:
       if (cret:=buffers.get(self)) is not None: return cret
-      buf = self.src[0].buffer
-      offset = self.src[1].val
+      if (cv := self.contiguous_view()) is None: raise RuntimeError(f"non-contiguous view is not supported for {self.device} buffer")
+      buf, offset = (b:=cv[0]).base.buffer, cv[1]
       if isinstance(buf, MultiBuffer):
         mbuf = MultiBuffer.__new__(MultiBuffer)
-        mbuf.bufs = [b.view(self.arg, self.dtype, offset * self.src[0].dtype.itemsize) for b in buf.bufs]
+        mbuf.bufs = [x.view(prod(self.max_shape), self.dtype, offset*b.dtype.itemsize) for x in buf.bufs]
         buffers[self] = mbuf
         return mbuf
-      assert isinstance(buf, Buffer), "must be a Buffer for SLICE"
-      buffers[self] = bv = buf.view(self.arg, self.dtype, offset * self.src[0].dtype.itemsize)
-      return bv
+      buffers[self] = buf.view(prod(self.max_shape), self.dtype, offset*b.dtype.itemsize)
+      return buffers[self]
     if self.op is Ops.MSELECT:
       ret = self.src[0].buffer
       assert isinstance(ret, MultiBuffer)
@@ -1181,7 +1164,7 @@ class UOp(RandMixin, metaclass=UOpMetaClass):
   def custom_function(name:str, *src:UOp) -> UOp: return UOp(Ops.CUSTOM_FUNCTION, src=src, arg=name)
 
   # opaque bodies stay as Ops.CALL; value-producing bodies become Ops.FUNCTION (wrapped in TUPLE)
-  _OPAQUE_CALL_BODIES = {Ops.SINK, Ops.PROGRAM, Ops.LINEAR, Ops.COPY, Ops.SLICE, Ops.CUSTOM_FUNCTION}
+  _OPAQUE_CALL_BODIES = {Ops.SINK, Ops.PROGRAM, Ops.LINEAR, Ops.COPY, Ops.CUSTOM_FUNCTION}
   def call(self, *srcs:UOp, ret_dtype:DType|None=None, grad_fxn:Callable|None=None,
            name:str|None=None, precompile:bool=False, precompile_backward:bool=False, aux:Any=None) -> UOp:
     if ret_dtype is not None: return UOp(Ops.CALL, ret_dtype, src=(self,)+srcs)
@@ -1775,10 +1758,14 @@ pm_unbind = PatternMatcher([(UPat(Ops.BIND, name="x"), do_unbind)])
 
 # ctx is source UOp for which we are finding a contiguous view for. used in contiguous_view_offset
 pm_contiguous_view_offset = PatternMatcher([
-  (UPat(Ops.INDEX, src=(UPat(),)), lambda: UOp.const(0)),
-  (UPat(Ops.INDEX, src=(UPat(), UPat(Ops.RANGE))), lambda: UOp.const(0)),
-  (UPat(Ops.INDEX, src=(UPat(), UPat(Ops.RANGE)+UPat.cvar('c'))), lambda c: c),
-  (UPat(Ops.INDEX, src=(UPat(), UPat.cvar('c'))), lambda ctx, c: c if resolve(ctx.numel() == 1, False) else None),
+  # normalize to 1d bitcasts
+  (UPat(Ops.BITCAST, name="b"), lambda b: b.src[0].flatten().bitcast(b.dtype).reshape(b.shape) if len(b.shape) != 1 else None),
+  (UPat(Ops.BITCAST, name="b").index(UPat.cvar("c")), lambda ctx, b, c:
+   b.src[0].flatten().index(UOp.range(ctx.numel() * (osz:=b.element_size())//(isz:=b.src[0].element_size()), 0) + (c * osz//isz)) if b.tag else None),
+  (UPat(Ops.INDEX, src=(UPat.var("b"),)), lambda b: b.rtag().index(0)),
+  (UPat(Ops.INDEX, src=(UPat.var("b"), UPat(Ops.RANGE))), lambda b: b.rtag().index(0)),
+  (UPat(Ops.INDEX, src=(UPat.var("b"), UPat(Ops.RANGE)+UPat.cvar('c'))), lambda ctx, b, c: b.rtag().index(c)),
+  (UPat(Ops.INDEX, src=(UPat.var("b"), UPat.cvar('c'))), lambda ctx, b, c: b.rtag().index(c) if resolve(ctx.numel() == 1, False) else None),
 ])
 
 # *** what was symbolic.py ***

tinygrad/uop/spec.py

@@ -233,13 +233,6 @@ spec_hcq = PatternMatcher([
 spec_full = PatternMatcher([
   (UPat(Ops.REWRITE_ERROR, dtypes.void, name="x"), lambda x: isinstance(x.arg, str)),
 
-  # SLICE on BUFFER is allowed if BUFFER is
-  (UPat(Ops.SLICE, src=(UPat(GroupOp.Movement.union({Ops.BUFFER, Ops.PARAM, Ops.STAGE, Ops.AFTER})),
-                        UPat(Ops.CONST, dtype=dtypes.weakint)), allow_any_len=True, name="bv"),
-   lambda bv: isinstance(bv.arg, int)),
-
-  (UPat(Ops.CALL, dtypes.void, src=(UPat((Ops.SLICE,)),), allow_any_len=True), lambda: True),
-
   # codegen may end ranges after gpudims has replaced RANGE with SPECIAL.
   (UPat(Ops.END, src=(UPat(), UPat()), allow_any_len=True), lambda: True),
 

tinygrad/uop/symbolic.py

@@ -293,8 +293,8 @@ symbolic = symbolic_simple+commutative+PatternMatcher([
   (UPat(Ops.AFTER, name="x"), lambda x: x.replace(src=(x.src[0],)+
     tuple(dedup(flatten([(y,) if y.op in {Ops.RANGE, Ops.STORE, Ops.CALL, Ops.FUNCTION, Ops.BARRIER, Ops.END, Ops.LINEAR, Ops.STAGE}
                         else y.src for y in x.src[1:]]))))),
-  # after with 1 src is just src[0]
-  (UPat(Ops.AFTER, src=(UPat.var("s"),)), lambda s: s),
+  # after/end with 1 src is just src[0]
+  (UPat((Ops.AFTER, Ops.END), src=(UPat.var("s"),)), lambda s: s),
 ])+div_and_mod_symbolic
 
 # ******** we take a small aside to "simplify_valid" to rewrite valids ********

tinygrad/viz/serve.py

@@ -50,7 +50,7 @@ uops_colors = {Ops.LOAD: "#ffc0c0", Ops.STORE: "#87CEEB", Ops.CONST: "#e0e0e0",
                Ops.INDEX: "#CEF9B7", Ops.STACK: "#D8F9E4",
                Ops.WMMA: "#efefc0", Ops.UNSHARD: "#f6ccff", Ops.INS: "#eec4ff",
                **{x:"#D8F9E4" for x in GroupOp.Movement}, **{x:"#ffffc0" for x in GroupOp.ALU}, Ops.THREEFRY:"#ffff80",
-               Ops.SLICE: "#E5EAFF", Ops.BUFFER: "#B0BDFF", Ops.GETADDR: "#9DB1F0", Ops.COPY: "#a040a0", Ops.CUSTOM_FUNCTION: "#bf71b6",
+               Ops.BUFFER: "#B0BDFF", Ops.GETADDR: "#9DB1F0", Ops.COPY: "#a040a0", Ops.CUSTOM_FUNCTION: "#bf71b6",
                Ops.CALL: "#00B7C8", Ops.FUNCTION: "#C07788", Ops.PARAM: "#14686F", Ops.SOURCE: "#c0c0c0", Ops.BINARY: "#404040",
                Ops.LINEAR: "#7DF4FF",
                Ops.ALLREDUCE: "#ff40a0", Ops.MSELECT: "#d040a0", Ops.MSTACK: "#d040a0", Ops.CONTIGUOUS: "#FFC14D",