geohot 0f4747fb5b custom kernels: Q4_K support (ggml type 12) ...

Q4_K is Q5_K without the high-bit array: same d/dmin/scales layout (so
_q5_scales works unchanged), 144-byte blocks, qs at word 4. both the dp4a
decode kernel and the WMMA prefill kernel take a ggml_type branch now.

Qwen3-8B Q4_K_M: decode 16.5 -> 114.8 tok/s, prefill 69 -> 536 tok/s

2026-08-23 17:20:20 -07:00

..

__init__.py

more test cleanups (#2631)

2023-12-05 16:17:57 -08:00

test_allreduce.py

use KernelCountException when kernel count is being compared (#17584)

2026-08-18 16:06:03 -07:00

test_assign.py

fix assign into bitcast with no explicit realize (#17531)

2026-08-14 09:23:01 -04:00

test_attention.py

minimize: master _attention with gated fused-scan swap, kernels/amd only, single chunk size

2026-08-22 13:50:47 -07:00

test_call.py

fix llm vars regression (kimi) (#17644)

2026-08-20 21:31:43 -07:00

test_callify.py

Tensor.callify will be the JIT (#14983)

2026-02-24 18:42:24 +08:00

test_conv.py

clean up some tests (#17154)

2026-07-23 11:22:18 -04:00

test_cpu.py

remove Ops.DEVICE from Ops.PROGRAM [PR] (#16797)

2026-06-29 17:41:31 -04:00

test_disk_tensor.py

fix Tensor(np) dtype and support fp8 safetensor (#17019)

2026-07-14 09:31:21 -04:00

test_dtype_spec.py

fix var of large half input (#17444)

2026-08-06 23:14:32 -04:00

test_dtype_weak.py

test update for weak const (#17675)

2026-08-21 21:50:29 -04:00

test_function.py

use KernelCountException when kernel count is being compared (#17584)

2026-08-18 16:06:03 -07:00

test_getitem_ops.py

addrspace cleanups (#16565)

2026-06-10 15:57:18 -07:00

test_gguf.py

move is_dtype_supported to renderer (#16226)

2026-05-20 21:19:37 -04:00

test_gradient.py

fix cast to weak twice [pr] (#17548)

2026-08-15 12:39:54 -04:00

test_hashing.py

some realize and corealize for slow tests (#17106)

2026-07-20 21:43:23 -04:00

test_hcq_graph.py

cpu hcq2 (#17503)

2026-08-14 15:06:03 +03:00

test_helpers.py

no weak Tensor from and to real buffer (#17067)

2026-07-17 16:09:10 -04:00

test_indexing.py

remove _force_unique from Tensor init (#16277)

2026-05-20 14:13:05 -04:00

test_invalid_tensor.py

UOp.is_invalid [PR] (#17351)

2026-08-01 02:36:32 -04:00

test_jit_cases.py

many jit tests belong in unit (#16508)

2026-06-04 21:36:53 -04:00

test_jit_footguns.py

many jit tests belong in unit (#16508)

2026-06-04 21:36:53 -04:00

test_jit.py

const(value, dtype) -> const(value).cast(dtype) in tests (#17335)

2026-07-31 13:24:31 -04:00

test_linalg.py

some realize and corealize for slow tests (#17106)

2026-07-20 21:43:23 -04:00

test_llm_amd.py

custom kernels: Q4_K support (ggml type 12)

2026-08-23 17:20:20 -07:00

test_llm_mla.py

refactor llm into files (#15780)

2026-04-17 12:33:11 +08:00

test_llm_moe.py

llm: respect expert_gating_func (#17458)

2026-08-17 12:19:33 -07:00

test_llm_server.py

generate: back to plain symbolic binding, the static-1 rule wasn't worth it

2026-08-23 16:54:54 -07:00

test_masked_tensor.py

delete test_tensor_uop (#13951)

2026-01-01 09:25:05 -05:00

test_metal_graph.py

remove Ops.SLICE (#17492)

2026-08-11 18:50:04 -04:00

test_multitensor.py

fix call arg indexing in shard scheduling (#17519)

2026-08-17 09:57:50 -07:00

test_objc.py

objc: fix _classmethods_ dispatch flag (#14854)

2026-04-20 09:35:03 +08:00

test_randomness.py

clean up some tests (#17154)

2026-07-23 11:22:18 -04:00

test_realize_is_realize.py

add device range as src[1] to multi (kimi) (#17264)

2026-07-28 16:39:34 -07:00

test_rearrange_einops.py

move more tests to test/null, split some existing ones (#14512)

2026-02-03 20:20:20 +08:00

test_schedule_cache.py

renumber invalids before callify (#17542)

2026-08-15 18:00:54 +09:00

test_setitem_schedule.py

add new schedule tests + format better (#17402)

2026-08-04 18:46:38 -07:00

test_shm_tensor.py

remove getenv(CI) (#16365)

2026-05-25 20:23:33 -04:00

test_symbolic_tensor.py

move more tests to test/null, split some existing ones (#14512)

2026-02-03 20:20:20 +08:00

test_system_pci_scan_bus.py

system: fix pci_scan_bus vendor filter (#14885)

2026-02-19 17:23:32 +03:00

test_tar.py

accept filename decorator [pr] (#8049)

2024-12-05 11:40:59 +08:00

test_tensor_data.py

const(value, dtype) -> const(value).cast(dtype) in tests (#17335)

2026-07-31 13:24:31 -04:00

test_tensor_io.py

move more tests to test/null, split some existing ones (#14512)

2026-02-03 20:20:20 +08:00

test_tinyfs.py

move fs_load and fs_store to nn (#16821)

2026-07-01 22:01:15 -04:00

test_winograd.py

move more tests to test/null, split some existing ones (#14512)

2026-02-03 20:20:20 +08:00