5 people 31e90ac370 [Attention] Enable FA4 sparse MQA decode on Blackwell (#55866) ...

Signed-off-by: Simon Veitner <sveitner@redhat.com>
Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>

2026-10-13 09:31:37 -04:00

..

attention

[Attention] Enable FA4 sparse MQA decode on Blackwell (#55866)

2026-10-13 09:31:37 -04:00

core

[Qwen4Exp] support Sequence Parallelism for Qwen3.8-Flash-Next (#56322)

2026-10-12 20:56:44 +08:00

fla

[Perf] Tune Triton GDN decode tiles on SM100-family GPUs (#61355)

2026-10-13 17:01:12 +08:00

helion

[MyPy][3/N] Fix MyPy errors in test groups (part 3) (#55939)

2026-09-30 11:06:07 +00:00

ir

[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

2026-09-16 10:57:24 -07:00

mamba

[Bugfix] Fix ReplaySSM ring shape assertion when TP does not divide Mamba2 n_groups (#60704)

2026-10-13 11:31:53 +00:00

mhc

[Test] Allow one bf16 ulp in the fused mHC post/pre residual check (#60724)

2026-10-09 17:54:06 -05:00

moe

[Bugfix] Fall back to auto for unquantized MoE layers when --moe-backend is marlin (#60219)

2026-10-13 08:48:13 +00:00

quantization

[Feature] Support batch invariance on ROCm (#52231)

2026-10-12 10:53:26 -05:00

scripts

[CI/Build][ROCm] Run the TileLang HIP symbol checks in their own interpreter (#53117)

2026-08-21 16:17:42 -05:00

turboquant

[Kernel][ROCm][Perf] FlyDSL decode-attention kernel for 4-bit TurboQuant KV cache (#47896)

2026-08-11 08:45:39 -07:00

ultraquant

[Attention] Add UltraQuant 4-bit KV cache backend (FlyDSL D=256) (#57057)

2026-10-06 20:42:52 +00:00

__init__.py

[CI/Build] Move test_utils.py to tests/utils.py (#4425)

2024-05-13 23:50:09 +09:00

allclose_default.py

Convert formatting to use ruff instead of yapf + isort (#26247)

2025-10-05 07:06:22 -07:00

conftest.py

[CI] Fix stale/fragile untethered kernels-root tests (#49423)

2026-07-22 14:43:07 -06:00

quant_utils.py

[Refactor] Remove dead tests utils (#58803)

2026-09-26 09:16:53 -04:00

test_apply_repetition_penalties.py

Add pydocstyle to the ruff rules (#52136)

2026-09-16 16:44:29 +00:00

test_awq_int4_to_int8.py

[CPU][Whisper] Support W4A16 quantized Whisper on the CPU WNA16 kernel (#58268)

2026-09-30 13:52:53 +08:00

test_bf16_skinny_gemm.py

[Misc] Consolidate csrc shared headers and prune dead code (#60721)

2026-10-12 09:46:22 -07:00

test_bf16x3_router_gemm_cutedsl.py

[Perf] Improve BF16x3 router GEMM accuracy and make it default on sm100 (#55899)

2026-09-09 07:33:12 -07:00

test_bf16x3_router_gemm_rocm.py

[ROCm][Perf] Use a stacked hipBLASLt GEMM for the bf16x3 router (#52668)

2026-10-05 17:28:07 +00:00

test_cache_kernels.py

Add pydocstyle to the ruff rules (#52136)

2026-09-16 16:44:29 +00:00

test_compressor_kv_cache.py

[Bugfix][DSv4] Don't bake the C128 compressor skip into CUDA graphs (#59709)

2026-10-13 10:15:11 +00:00

test_concat_mla_q.py

[Bugfix][ROCm] Use the platform FP8 range in the concat MLA q test (#58153)

2026-09-22 18:05:33 -05:00

test_cp_gather_fp8.py

[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

2026-09-16 10:57:24 -07:00

test_cpu_fused_sampling.py

[Bugfix][CPU] Fix fused Gumbel-max wrapping-window bias (#60221)

2026-10-08 15:15:53 +08:00

test_deepseek_v4_cpu_kernels.py

[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

2026-09-16 10:57:24 -07:00

test_dsv41_mega_attn_layouts.py

[Model][DeepSeek-V4.1] Permute mega-attention wq_b / wo_a while loading (#60064)

2026-10-06 05:20:09 +00:00

test_engram.py

[Perf][DSv4.1] Exchange packed FP8 Engram rows across DP and feed wkv MXFP8 (#59923)

2026-10-09 15:30:18 +08:00

test_flashinfer_bf16_gemm.py

[Kernel] Remove WAR for LL CuTeDSL BF16 GEMM (#60434)

2026-10-08 05:38:07 +00:00

test_flex_attention.py

Add pydocstyle to the ruff rules (#52136)

2026-09-16 16:44:29 +00:00

test_fp32_router_gemm.py

[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

2026-09-16 10:57:24 -07:00

test_fused_deepseek_v4_qnorm_rope_kv_insert.py

[CI][ROCm] Add ten AMD parity groups (#57876)

2026-09-21 11:08:06 +00:00

test_fused_deepseek_v32_norm_rope.py

[Kernel][Build] Bump DeepGEMM to the 26/09/30 release and adapt the sparse indexer (#61149)

2026-10-12 23:24:43 -07:00

test_fused_indexer_q_rope_quant.py

[Kernel][Build] Bump DeepGEMM to the 26/09/30 release and adapt the sparse indexer (#61149)

2026-10-12 23:24:43 -07:00

test_fused_inv_rope_fp8_quant.py

[Perf][DSv4.1] Fuse small-batch WO-A with inverse RoPE and MXFP8 quant on SM100/SM103 (#58634)

2026-09-26 21:16:12 -07:00

test_fused_minimax_m3_qknorm_rope_kv_insert.py

[Attention][MiniMax-M3] NVFP4 KV cache on the MSA sparse attention path (#59300)

2026-10-01 14:39:55 -07:00

test_fused_qk_norm_rope_gate.py

[Kernel][Perf] Enable fused QK-norm + partial MRoPE + gate for Qwen3.6 (#52676)

2026-08-24 18:28:49 -07:00

test_fused_quant_activation.py

[Fusion] Support 3D activation quantization fusion and enable it for Pixtral (#59612)

2026-10-09 15:12:03 +08:00

test_gate_linear_rocm_dispatch.py

[ROCm][Perf] Use a stacked hipBLASLt GEMM for the bf16x3 router (#52668)

2026-10-05 17:28:07 +00:00

test_gemm_rs_ar.py

[Kernel][DSV4.1] Fuse MXFP8 wo_b GEMM with sequence-parallel reduce-scatter (#57428)

2026-09-22 01:07:17 -07:00

test_kpool_decode_update_batched.py

[Kernel][Build] Bump DeepGEMM to the 26/09/30 release and adapt the sparse indexer (#61149)

2026-10-12 23:24:43 -07:00

test_ll_bf16_gemm.py

[Kimi-K3][Perf] Prefetch ll_bf16 router weights for M=1 (#53524)

2026-09-01 21:40:19 +00:00

test_minimax_m3_amd_ops.py

[Bugfix][Quantization] Fix MXFP8 startup crash on layers below mm_mxfp8 shape limits (#54223)

2026-09-25 02:54:11 +00:00

test_minimax_m3_sparse_attn_fp8_scale.py

[ROCm][MiniMax-M3] Add AITER sparse paged attention (#47287)

2026-07-12 19:27:29 -07:00

test_onednn.py

[CPU] Speedup LM Head on Arm CPUs (#55352)

2026-09-11 16:21:14 +08:00

test_relu2_fp8_quant.py

[Perf][Kernel][Quantization] Fuse ReLU2 with static FP8 activation quantization (#53793)

2026-09-14 10:26:15 -07:00

test_rocm_fp32_router_gemm.py

[ROCm][Perf] Add low-M FP32 router GEMM for gfx950 (#54845)

2026-09-03 10:42:25 -07:00

test_rocm_gemm_rs_ar.py

[ROCm][CI] Mirror generic GEMM-RS/AR on MI355 (#58433)

2026-09-29 10:28:15 +00:00

test_shuffle_rows.py

[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

2026-09-16 10:57:24 -07:00

test_top_k_per_row.py

[ROCm][Perf][GLM-5.3-Flash] Add AITER topk backend for decodes (#58167)

2026-10-02 11:47:41 +01:00

utils.py

[Refactor] Remove dead tests utils (#58803)

2026-09-26 09:16:53 -04:00