qazal and GitHub
04c271ac41
simplify digitalocean_mi350x ( #17502 )
...
* simplify digitalocean_mi350x
* no hardcoded rocm path
2026-08-12 16:00:04 +09:00
qazal and GitHub
e1013a6356
llama: create dataset cache by default in dev_beam ( #17500 )
2026-08-12 15:03:39 +09:00
wozeparrot and GitHub
f891f5ffd0
gptoss: route lm_head thru asm_gemm ( #17497 )
2026-08-11 18:32:50 -07:00
qazal and GitHub
22722ea2e7
llama: correct optim_dtype for mxfp4 ( #17482 )
Autogen / In-tree Autogen (comgr 2) (push) Successful in 45s
Unit Tests / Torch Backend Tests (push) Failing after 56s
Unit Tests / Linters (push) Failing after 1m25s
Unit Tests / Python Backend (push) Failing after 3m5s
Unit Tests / Fuzzing (push) Failing after 23s
Unit Tests / Docs (push) Successful in 3m32s
Unit Tests / Null Tests (push) Successful in 3m59s
Unit Tests / CL IMAGE Tests (push) Failing after 43s
Unit Tests / Unit Tests (push) Successful in 3m32s
Unit Tests / openpilot Compile Tests (push) Failing after 46s
Unit Tests / Test LLM (push) Failing after 15s
Deploy Docs / deploy (push) Successful in 4m35s
Unit Tests / SPEC=2 (1) (push) Successful in 3m49s
Unit Tests / Optimization Tests (push) Failing after 48s
Unit Tests / Linux (DEV=CL) (push) Failing after 36s
Unit Tests / SPEC=2 (2) (push) Successful in 3m50s
Unit Tests / Linux (DEV=CPU:X86) (push) Failing after 23s
Unit Tests / Linux (DSP) (push) Successful in 1m41s
Unit Tests / Models (push) Successful in 3m30s
Unit Tests / Linux (DEV=CPU:LVP) (push) Successful in 3m8s
Unit Tests / Linux (DEV=CPU:LLVM) (push) Successful in 3m20s
Unit Tests / hcq2 (push) Failing after 50s
Unit Tests / AMD ASM IDE (push) Failing after 2m45s
Unit Tests / Linux (DEV=CPU:CLANG) (push) Successful in 4m25s
Unit Tests / Linux (am) (push) Failing after 55s
Unit Tests / ONNX (CPU) Tests (push) Successful in 5m28s
Autogen / In-tree Autogen (push) Successful in 10m45s
Unit Tests / Linux (DEV=WEBGPU) (push) Failing after 5m4s
Unit Tests / Linux (amdllvm gfx1100) (push) Successful in 3m51s
Unit Tests / Linux (amd gfx1100) (push) Successful in 4m23s
Unit Tests / Compile-only (ir3) (push) Failing after 28s
Unit Tests / Linux (amdllvm gfx1201) (push) Successful in 3m53s
Unit Tests / Compile-only (nak) (push) Failing after 28s
Unit Tests / Linux (amd gfx1201) (push) Failing after 4m33s
Unit Tests / Linux (amdllvm gfx950) (push) Successful in 4m5s
Unit Tests / Linux (ptx) (push) Failing after 3m24s
Unit Tests / Linux (amd gfx950) (push) Successful in 5m55s
Unit Tests / Linux (nv) (push) Failing after 5m17s
Autogen / In-tree Autogen (macos) (push) Canceled after 0s
Benchmarks / Mac pytest (push) Canceled after 0s
Benchmarks / LLM (DEV=AMD) (push) Canceled after 0s
Benchmarks / LLM (DEV=METAL) (push) Canceled after 0s
Benchmarks / LLM (DEV=NV) (push) Canceled after 0s
Benchmarks / HLB-CIFAR10 (DEV=AMD) (push) Canceled after 0s
Benchmarks / HLB-CIFAR10 (DEV=METAL) (push) Canceled after 0s
Benchmarks / HLB-CIFAR10 (DEV=NV) (push) Canceled after 0s
Benchmarks / MLPerf (AMD) (push) Canceled after 0s
Benchmarks / MLPerf (NV) (push) Canceled after 0s
Benchmarks / Stable Diffusion (DEV=AMD) (push) Canceled after 0s
Benchmarks / Stable Diffusion (DEV=METAL) (push) Canceled after 0s
Benchmarks / Stable Diffusion (DEV=NV) (push) Canceled after 0s
Benchmarks / Tests (DEV=AMD) (push) Canceled after 0s
Benchmarks / Tests (DEV=METAL) (push) Canceled after 0s
Benchmarks / Tests (DEV=NV) (push) Canceled after 0s
Benchmarks / UsbGPU Benchmark (push) Canceled after 0s
Benchmarks / comma Benchmark (0.11.2) (push) Canceled after 0s
Benchmarks / comma Benchmark (0.11.0) (push) Canceled after 0s
Benchmarks / DSP Benchmark (push) Canceled after 0s
Benchmarks / UsbGPU Benchmark (comma) (push) Canceled after 0s
Benchmarks / PCI Driver Benchmark (DEV=AMD) (push) Canceled after 0s
Benchmarks / PCI Driver Benchmark (DEV=NV) (push) Canceled after 0s
Benchmarks / LLVM Speed (push) Canceled after 0s
Platform Tests / MacOS (unit) (push) Canceled after 0s
Platform Tests / MacOS (unit, mock) (push) Canceled after 0s
Platform Tests / MacOS (DEV=METAL) (1) (push) Canceled after 0s
Platform Tests / MacOS (DEV=METAL) (2) (push) Canceled after 0s
Platform Tests / MacOS (DEV=CPU:CLANG) (push) Canceled after 0s
Platform Tests / MacOS (DEV=CPU:LLVM) (push) Canceled after 0s
Platform Tests / MacOS (DEV=CPU:LVP) (push) Canceled after 0s
Platform Tests / MacOS (DEV=WEBGPU) (push) Canceled after 0s
Platform Tests / Windows (DEV=CPU:CLANG) (push) Canceled after 0s
Platform Tests / Windows (DEV=CPU:LLVM) (push) Canceled after 0s
Platform Tests / Windows (DEV=CPU:X86) (push) Canceled after 0s
Platform Tests / Windows (DEV=WEBGPU) (push) Canceled after 0s
Platform Tests / Compile-only (QCOM CL) (push) Canceled after 0s
2026-08-11 03:01:28 +09:00
qazal and GitHub
2821bd646f
late loss.to("CPU") in llama ( #17476 )
...
* late loss.to("CPU") in llama
* acc = 0
2026-08-10 17:31:50 +09:00
qazal and GitHub
44f1f45cd5
llama: custom silu kernels ( #17462 )
...
* start by copying the C
* uop kernel
* cleanup tests
* estimates is part of SPEC
2026-08-10 16:43:01 +09:00
qazal and GitHub
9dd3b8402e
default llama 8b to MXFP4=1 ( #17465 )
2026-08-09 00:13:32 +08:00
qazal and GitHub
9636dd1a25
test MXFP4 llama without hipcc ( #17435 )
...
* test MXFP4 llama without hipcc
* first pythonpath then dev
2026-08-06 17:31:40 +09:00
qazal and GitHub
f993228b73
llama: accurate mxfp4 mfu ( #17388 )
...
* llama: accurate mxfp4 mfu
* train_llama3 import
2026-08-04 18:41:57 +09:00
qazal and GitHub
be5f62d269
llama: refactor amax stuff and skip in fp4 ( #17375 )
2026-08-03 20:05:52 +09:00
wozeparrot and GitHub
e14cadb1fb
gptoss: set ASM_GEMM ( #17363 )
2026-08-02 06:52:36 -07:00
wozeparrot and GitHub
98b700bad1
gptoss optim fixes ( #17356 )
2026-08-01 09:10:06 -07:00
George Hotz and GitHub
a88f832f0c
remove UOp.val ( #17345 )
2026-07-31 18:58:38 -07:00
wozeparrot and GitHub
4f5cadd15d
gptoss ci ( #17325 )
2026-07-31 05:51:58 -07:00
wozeparrot and GitHub
93b74c75fc
gptoss: grouped moe ( #17322 )
2026-07-31 03:25:32 -07:00
qazal and GitHub
f7964acb64
llama with MXFP4 ( #17321 )
...
* mxfp4 in llama
* less
* name
2026-07-31 18:25:25 +09:00
sirhcm and GitHub
13452b3775
benchmark comma big model ( #17312 )
2026-07-31 01:37:34 -04:00
wozeparrot and GitHub
b5a2a5666a
gptoss moe routing ( #17284 )
2026-07-30 07:49:14 -07:00
sirhcm and GitHub
bfc9fc6e0a
nicer TinyJit decorator ( #17293 )
2026-07-29 21:48:00 -04:00
George Hotz and GitHub
57ae1bc7a7
rename MULTI to UNSHARD ( #17267 )
...
* rename MULTI to UNSHARD
* comment updates (glm)
* rename method to unshard
2026-07-28 16:51:41 -07:00
chenyu and GitHub
23e9e76e8c
DEFAULT_FLOAT/DEFAULT_INT ContextVar [pr] ( #17265 )
2026-07-28 19:08:18 -04:00
755dfb243b
rename CPU_COUNT to NUM_CPU_THREADS with cgroup awareness ( #17263 )
...
Rename CPU_COUNT to NUM_CPU_THREADS so it can be overridden via env var.
Default uses _get_cpu_count() which respects cgroup limits:
- os.process_cpu_count() on Python 3.13+
- /sys/fs/cgroup/cpu.max on cgroup v2
- /sys/fs/cgroup/cpu/cpu.cfs_quota_us on cgroup v1
- os.sched_getaffinity(0) fallback
Use NUM_CPU_THREADS.value in the dataloader instead of cpu_count(),
and update export_model.py and all renderer references.
Co-authored-by: teeny-runner <runner@teeny>
2026-07-28 15:33:57 -07:00
wozeparrot and GitHub
056974468e
gptoss: split no-wd params ( #17233 )
2026-07-27 02:58:43 -07:00
wozeparrot and GitHub
f0117e98df
refactor mlperf optim ( #17200 )
2026-07-25 10:37:11 -07:00
wozeparrot and GitHub
2e7db75668
gptoss: faster swa ( #17150 )
2026-07-23 06:56:27 -07:00
wozeparrot and GitHub
5f2eaeee40
gptoss: rope freqs in fp32 ( #17132 )
2026-07-22 06:11:06 -07:00
qazal and GitHub
1cf8f2f68c
llama: inplace amax update ( #17064 )
...
* llama: inplace amax update
* remove amax_out return
* work
* fit
* work
* work
* keep
* diff cleanup
2026-07-20 15:05:41 +09:00
wozeparrot and GitHub
ee290b3e39
optim: mxfp8 zero 1 allgathers in fp8 ( #17073 )
2026-07-18 07:44:50 -07:00
wozeparrot and GitHub
a836c3822a
gptoss: 3d mx block scale ( #17062 )
2026-07-16 23:30:24 -07:00
wozeparrot and GitHub
0d30f97584
mlperf: make v6.1 dir ( #17031 )
2026-07-15 10:40:55 -07:00
wozeparrot and GitHub
06169f5013
gptoss: small fixes ( #17025 )
2026-07-14 20:40:23 -07:00
qazal and GitHub
939f28d571
fused qkv rope custom kernel ( #17021 )
...
* work
* fused qkv_norm
* work
* speed
* not that yet
* test cleanup
* just clone
* remove .realize()
* cleanup tests
2026-07-15 01:08:42 +09:00
qazal and GitHub
edfef062ed
skip viz.cli -t in null device ( #17018 )
2026-07-14 19:24:58 +09:00
qazal and GitHub
cae6696d75
llama: split current and next amax state ( #16993 )
2026-07-12 18:52:33 +09:00
George Hotz and GitHub
40de90ab19
lil changes to cifar ( #16972 )
...
* lil changes to cifar
* lil changes
* pool
2026-07-10 14:19:46 -07:00
George Hotz and GitHub
77823056d4
remove dead DSP code and pre_matcher ( #16968 )
2026-07-10 11:27:50 -07:00
wozeparrot and GitHub
5cc31e23e6
gptoss: scripts ( #16962 )
2026-07-09 20:50:20 -07:00
George Hotz and GitHub
7f8bbe5407
hlb_cifar free speed ( #16938 )
...
* free speed for hlb cifar
* mirror
2026-07-08 15:49:48 -07:00
George Hotz and GitHub
fdffc6c0c8
remove dtypes base ( #16931 )
...
* remove dtypes base
* find/replace bug
* vcount is junk too
2026-07-08 11:48:11 -07:00
wozeparrot and GitHub
ab131c2086
gptoss: zero-1 optim ( #16916 )
2026-07-07 18:34:47 -07:00
Divij and GitHub
9f388d42b7
padded maxpool with -inf as done in pytorch ( #16858 )
2026-07-07 16:10:44 -07:00
wozeparrot and GitHub
a805ce03b1
gptoss: model train ( #16884 )
2026-07-06 22:07:50 -07:00
wozeparrot and GitHub
a330bfffc9
gptoss: remove contiguous ( #16882 )
2026-07-05 17:48:32 -07:00
wozeparrot and GitHub
35522af0df
gptoss: use quantize_mxfp8 ( #16881 )
2026-07-05 17:48:24 -07:00
George Hotz and GitHub
3fd6f3d28c
no more GEP ( #16866 )
2026-07-04 13:42:28 -07:00
qazal and GitHub
07f7383d29
llama: remove unused bf16 assembly gemm ( #16859 )
...
* only hk bf16 gemm
* rm asm gemm
* more cleanup
* half isn't supported in asm gemm anymore
* more test edits
* unused
* remove TestMagicGu
* uop gemm is still tested
* minimal diff
2026-07-04 18:41:54 +09:00
wozeparrot and GitHub
42be177ea8
gpt-oss-20b: model ( #16837 )
...
* feat: gpt-oss model file
* clean: don't need that function
* feat: working model
2026-07-03 01:29:56 -07:00
chenyu and GitHub
3a9b87e48f
remove Ops.DEVICE from Ops.PROGRAM [PR] ( #16797 )
2026-06-29 17:41:31 -04:00
chenyu and GitHub
36c0e81979
deprecate Tensor.training [PR] ( #16774 )
2026-06-27 11:02:28 -04:00
George Hotz and GitHub
a94a32ff71
move image to post coalese ( #16749 )
...
* move image to post coalese
* fix types
* dv2
* work
* load_store_indexing
* lsi
* simplify indexing
* always simplify
* more
* whitespace
* no simplify
* no early opt
* fix linter
* exit early without valid
* always simplify
* relax ir3
* no gep
* Revert "no gep"
This reverts commit 3fa4dd6dfa
2026-06-25 23:33:07 -07:00
sirhcmandGitHub