3 people 7a2fdbaac4 [K3 Perf] Fuse MXFP4 top-k finalization into latent-tail, ~5% E2E latency reduction (#53152) ...

Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
Co-authored-by: Yongye Zhu <zyy1102000@gmail.com>
Co-authored-by: OpenAI Codex <codex@openai.com>

2026-08-21 13:29:59 -04:00

..

cpu

[Hardware][Power] Add FAST_EXP for Power (#49571)

2026-07-27 05:47:29 +00:00

deepgemm

Add platform manual_seed_all API (#38468)

2026-04-10 13:43:50 +08:00

ir

[vLLM IR] Add IR op testing and benchmarking infrastructure (#40167)

2026-04-21 00:23:03 +00:00

__init__.py

[vLLM IR] Add IR op testing and benchmarking infrastructure (#40167)

2026-04-21 00:23:03 +00:00

bench_concat_mla_q.py

[Attention][Perf][Kernel] Replace torch.cat with vectorized CUDA kernel MLA query concat - DeepSeek-V3.2 (#34917)

2026-03-09 09:50:36 -07:00

bench_cp_gather_fp8.py

[Bugfix] Sparse MLA: enable fp8_ds_mla dense prefill (#48642)

2026-07-16 22:44:04 +00:00

benchmark_2d_silu_mul_fp8_quant.py

[Bugfix][CI] fix typos (#34934)

2026-03-05 17:05:46 +00:00

benchmark_activation.py

[Bugfix] Fix kernel benchmark (#33752)

2026-02-22 21:18:08 -08:00

benchmark_block_fp8_gemm.py

[W8A8 Block Linear Refactor][2/N] Remove W8A8Fp8BlockLinearOp and adopt Fp8 block linear kernel selections. (#33892)

2026-04-09 08:50:39 +08:00

benchmark_cp_gather.py

[Kernel] Optimize long-context MLA cache gathers (#51739)

2026-08-10 22:23:43 -07:00

benchmark_cutlass_moe_fp8.py

[MoE] Move cutlass moe to fused_moe/experts/ (#40574)

2026-04-24 06:05:49 +00:00

benchmark_cutlass_moe_nvfp4.py

[MoE] Move cutlass moe to fused_moe/experts/ (#40574)

2026-04-24 06:05:49 +00:00

benchmark_device_communicators.py

[Hardware] Replace torch.cuda.device_count/current_device/set_device API (#36145)

2026-03-12 07:57:47 -07:00

benchmark_flydsl_moe_w4a16.py

[CI/Build][BugFix][The Rock] Fix get_ssm_device_name to return sanitized, usable filename (#47781)

2026-07-08 12:12:54 +00:00

benchmark_fp8_gemm.py

[Bugfix] Fix kernel benchmark (#33752)

2026-02-22 21:18:08 -08:00

benchmark_fused_collective.py

[Perf] Enable + tune FlashInfer fused allreduce at world_size=16 on SM 10.3 (GB300) (#46392)

2026-06-24 23:04:17 -07:00

benchmark_fused_moe_lora_one_shot.py

[LoRA] Add one shot triton kernel For MoE LoRA (#42290)

2026-05-25 19:47:04 -07:00

benchmark_fused_q_cutedsl.py

[Kernel][SM100] Add a CuTeDSL fused query kernel (#49792)

2026-08-04 15:23:06 -07:00

benchmark_fused_topk.py

[Kernel] Add topk_sigmoid kernel (#31246)

2026-01-21 22:49:51 +00:00

benchmark_grouped_gemm_cutlass.py

[MoE] Move cutlass moe to fused_moe/experts/ (#40574)

2026-04-24 06:05:49 +00:00

benchmark_inkling_qkvr_prep.py

[ROCm] [Model] Enable TML inkling (#48841)

2026-07-27 10:05:46 +00:00

benchmark_int8_gemm.py

[Bugfix] Fix kernel benchmark (#33752)

2026-02-22 21:18:08 -08:00

benchmark_k3_cutedsl_residual.py

[Model] Add Kimi K3 support: model files and kernels [1/N] (#50089)

2026-07-29 14:10:58 +08:00

benchmark_kimi_k3_gemm_rs_ar.py

[Kimi-K3] Extend GEMM-RS to GEMM-AR (#53053)

2026-08-21 03:44:15 +00:00

benchmark_kimi_k3_kda_decode.py

[ROCm][Perf] Kimi-K3 Fused kernel for KDA decode (#50654)

2026-08-11 20:39:34 -07:00

benchmark_kimi_k3_latent_moe_tail.py

[K3 Perf] Fuse MXFP4 top-k finalization into latent-tail, ~5% E2E latency reduction (#53152)

2026-08-21 13:29:59 -04:00

benchmark_kimi_k3_sp_collectives.py

[Model] Add Kimi K3 support: model files and kernels [1/N] (#50089)

2026-07-29 14:10:58 +08:00

benchmark_layernorm.py

[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

2026-03-05 10:36:39 +00:00

benchmark_lora.py

[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

2026-03-05 10:36:39 +00:00

benchmark_machete.py

[Refactor] Rename gptq_marlin to marlin to match MoE (#32952)

2026-01-23 16:48:12 -05:00

benchmark_marlin.py

[Quantization][Deprecation] Remove Marlin 24 (#32688)

2026-01-28 15:54:59 +00:00

benchmark_mla_k_concat.py

[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

2026-03-05 10:36:39 +00:00

benchmark_moe_align_block_size.py

Add platform manual_seed_all API (#38468)

2026-04-10 13:43:50 +08:00

benchmark_moe_defaults.py

Revert "[Platform] Replace torch.cuda.Event with torch.Event (#47140)" (#47668)

2026-07-06 14:18:33 +01:00

benchmark_moe_permute_unpermute.py

[Perf] Optimize moe permute by pre-allocate buffer, 9~14% kernel performance improvement (#43014)

2026-05-28 06:18:26 -07:00

benchmark_moe.py

[Bugfix] Add Kimi K3 MoE support to benchmark_moe.py (#50082)

2026-08-19 19:42:30 +00:00

benchmark_mrope.py

[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

2026-03-05 10:36:39 +00:00

benchmark_mxfp4_qutlass.py

[Bugfix] Fix kernel benchmark (#33752)

2026-02-22 21:18:08 -08:00

benchmark_nvfp4_gemm.py

[Bugfix] Fix kernel benchmark (#33752)

2026-02-22 21:18:08 -08:00

benchmark_nvfp4_quant.py

[Bugfix] Fix kernel benchmark (#33752)

2026-02-22 21:18:08 -08:00

benchmark_nvfp4_qutlass.py

[Bugfix] Fix kernel benchmark (#33752)

2026-02-22 21:18:08 -08:00

benchmark_paged_attention.py

Delete PagedAttention (#47361)

2026-07-02 12:31:26 -07:00

benchmark_per_token_group_quant.py

[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

2026-03-05 10:36:39 +00:00

benchmark_per_token_quant_fp8.py

[Bugfix] Fix kernel benchmark (#33752)

2026-02-22 21:18:08 -08:00

benchmark_quant.py

[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

2026-03-05 10:36:39 +00:00

benchmark_rdna_hybrid_w4a16_gemm.py

[ROCm][Kernel] Add HybridW4A16LinearKernel: Triton prefill + HIP skinny decode (#40977)

2026-07-13 20:22:00 -07:00

benchmark_relu_squared.py

[Kernel] Implement CUDA kernel for ReLUSquaredActivation (relu^2) (#39058)

2026-07-12 19:18:03 -07:00

benchmark_reshape_and_cache_flash.py

[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

2026-03-05 10:36:39 +00:00

benchmark_reshape_and_cache.py

[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

2026-03-05 10:36:39 +00:00

benchmark_rmsnorm.py

Update Optional[x] -> x | None and Union[x, y] to x | y (#26633)

2025-10-12 09:51:31 -07:00

benchmark_rope.py

[Bugfix] Fix kernel benchmark (#33752)

2026-02-22 21:18:08 -08:00

benchmark_router_gemm.py

[PERF]MiniMax-M2 gate kernel (#38445)

2026-05-29 18:28:34 -07:00

benchmark_selective_state_update.py

[XPU] Add tuned Mamba SSU configs for Intel Arc Pro B70 (#50534)

2026-08-13 12:36:55 +00:00

benchmark_shapes.py

Disable nm-testing models with issues in CI (#28206)

2025-11-06 06:19:07 -08:00

benchmark_silu_mul_fp8_quant.py

[MoE] Move DEEP_GEMM into experts/ subdirectory (#39005)

2026-04-08 19:23:08 +00:00

benchmark_thinking_budget.py

[Feat] Support thinking_token_budget in Model Runner V2 (#46727)

2026-08-07 03:36:18 +00:00

benchmark_trtllm_decode_attention.py

[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

2026-03-05 10:36:39 +00:00

benchmark_trtllm_prefill_attention.py

[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize (#36085)

2026-03-05 10:36:39 +00:00

benchmark_vit_aiter_fp8_attn.py

[ROCm] Add AITER FP8 ViT encoder attention (#49937)

2026-07-30 06:22:10 +00:00

benchmark_vit_bilinear_pos_embed.py

[Perf] triton bilinear_pos_embed kernel for ViT (#37948)

2026-04-01 01:52:02 -07:00

benchmark_vit_fp8_attn.py

[Perf] FP8 FlashInfer Attn for ViT (#38065)

2026-04-27 13:44:15 +08:00

benchmark_w8a8_block_fp8.py

[CI/Build][BugFix][The Rock] Fix get_ssm_device_name to return sanitized, usable filename (#47781)

2026-07-08 12:12:54 +00:00

graph_machete_bench.py

[Chore]:Extract math and argparse utilities to separate modules (#27188)

2025-10-26 04:03:32 -07:00

requirements.txt

[Kernel] (2/N) Machete - Integrate into CompressedTensorsWNA16 and GPTQMarlin (#7701)

2024-09-23 13:46:26 -04:00

utils.py

Update Optional[x] -> x | None and Union[x, y] to x | y (#26633)

2025-10-12 09:51:31 -07:00

weight_shapes.py

[kernel] Support W4A8 on Hopper (#23198)

2025-08-24 06:18:04 +00:00