3 people 9d14fd3095 [Kernel][MoE] DeepEP v2: async finalize to overlap shared experts with combine (#52781) ...

Signed-off-by: Mario <404mario@users.noreply.github.com>
Co-authored-by: Mario <404mario@users.noreply.github.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

2026-09-15 00:04:22 -04:00

..

modular_kernel_tools

[Log] Update deepgemm log (#45857)

2026-06-17 15:34:22 +00:00

__init__.py

[Kernel] DeepEP dispatch-combine kernel integration (#18434)

2025-06-03 12:30:02 -07:00

conftest.py

[CI] Restore clean GPU state before OAI Triton MoE tests (#55453)

2026-09-08 01:15:21 -07:00

parallel_utils.py

[ROCm][Perf][Bugfix] Multi-stream perf improvements; rocprofiler fixes (#55099)

2026-09-08 17:42:34 -07:00

test_b12x.py

[Bugfix] DSv4 MXFP4 selector: stop narrowing explicit aliases to their BF16 variant (#53586)

2026-09-07 07:19:13 +00:00

test_batched_deepgemm.py

[MoE] Remove inplace fused experts mechanism (#43727)

2026-05-27 20:00:19 -07:00

test_batched_moe.py

[MoE] Generalize masked activation for padded layouts (#51217)

2026-09-01 11:34:13 -04:00

test_block_fp8.py

[Model] Support Qwen3.8-Flash-Next (#53896)

2026-08-31 05:57:56 +00:00

test_block_int8.py

[CI][Bugfix] Fix test isolation in block_int8/ptpc_fp8 MoE kernel tests (#49609)

2026-07-23 12:26:02 -05:00

test_count_expert_num_tokens.py

[XPU][Tests] Make tests device-agnostic (#51968)

2026-08-20 02:25:41 -07:00

test_cpu_fused_moe.py

[Model] Add DeepSeek-V4 CPU backend (#55355)

2026-09-10 14:14:48 +08:00

test_cpu_int4_moe.py

[CPU][BugFix] Multiple fixes to w4a8_int8 CPU MoE path (#46739)

2026-07-06 05:39:20 +00:00

test_cpu_quant_fused_moe.py

[Model] Add DeepSeek-V4 CPU backend (#55355)

2026-09-10 14:14:48 +08:00

test_cutedsl_moe.py

[Bugfix] Gate sm_100-only kernel tests on the capability family, not >= (#54306)

2026-09-01 11:29:15 -04:00

test_cutlass_moe.py

[Bugfix] Handle padded routes in CUTLASS MoE permutations (#54747)

2026-09-01 19:08:17 -07:00

test_deepep_deepgemm_moe.py

use split_group for pytorch process group creation (#41980)

2026-06-04 14:36:07 -04:00

test_deepep_moe.py

[ROCm][Perf][Bugfix] Multi-stream perf improvements; rocprofiler fixes (#55099)

2026-09-08 17:42:34 -07:00

test_deepep_v2_async_finalize.py

[Kernel][MoE] DeepEP v2: async finalize to overlap shared experts with combine (#52781)

2026-09-15 00:04:22 -04:00

test_deepep_v2_moe.py

[Bugfix] vLLM crashes at startup when DeepEP v2 is used with --enforce-eager wiht TRTLLM Bf16 (#51824)

2026-08-19 20:18:47 -04:00

test_deepgemm.py

[Kimi K3 Bug] Fix deepgemm support for kimi k3 (#50458)

2026-07-30 20:10:16 -04:00

test_flashinfer_b12x_moe.py

[BugFix] weights processing peak memory reduction for nvfp4 MoE layers (#46276)

2026-07-11 02:05:35 +00:00

test_flashinfer_cutedsl_layout.py

[Kernel] Support Nvfp4 Cutedsl Moe Swiglu-oai and Relu2(non-gated) Activation (#47106)

2026-08-06 17:42:45 -07:00

test_flashinfer_cutedsl_nvfp4_moe.py

feat: extended EPLB support for Mistral Large 3 and additional MoE backends (#48355)

2026-08-07 05:32:52 -07:00

test_flashinfer_moe.py

Revert "[Bugfix][MoE] Tune FlashInfer experts to scheduler token limit" (#52989) (#53186)

2026-08-21 12:39:28 -06:00

test_flashinfer.py

[Bugfix][MoE] Allow TRTLLM FP8 block-scale MoE with SwiGLU clamp (#55069)

2026-09-04 08:34:21 -07:00

test_flydsl_moe.py

[ROCm][CI] Stabilize MI355 FlyDSL MoE accuracy test (#53024)

2026-08-21 13:17:18 -07:00

test_fused_topk.py

[CI][MoE] Moe kernels test cleanup (#54954)

2026-09-02 13:31:36 -05:00

test_gemma4router.py

[FEAT] [Perf] [Gemma4] Fused Gemma4 Routing Function Triton (#39083)

2026-04-19 09:57:39 +00:00

test_gpt_oss_triton_kernels.py

[Perf] Reuse topk SparseMatrix routing metadata in GPT-OSS MoE forward (#45457)

2026-08-29 12:14:39 -04:00

test_grouped_topk.py

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

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

test_marlin_vs_trtllm_mxint4.py

[Quant][Kernel] Remove GPTQ Group/Dynamic Activation Ordering (#54809)

2026-09-08 12:45:12 -07:00

test_modular_kernel_combinations.py

[MoE Refactor] Remove supports_expert_map (#43108)

2026-05-29 17:26:56 -04:00

test_modular_oai_triton_moe.py

[ROCm][CI] Enable modular OAI Triton MoE tests (#46434)

2026-08-19 13:46:12 -05:00

test_moe_align_block_size.py

[Kernel][MoE] Optimize batched_moe_align_block_size with cooperative writes (#53280)

2026-09-11 12:05:22 -07:00

test_moe_fused_mul_sum.py

[Quantization][Humming] Support MXFP4 weight + block-FP8 activation for MoE (#51332)

2026-08-25 10:21:00 -04:00

test_moe_kernel_oracle.py

[MoE] [MoE Refactor] Add moe kernel oracle abc 37753 (#43461)

2026-06-25 09:34:03 -04:00

test_moe_layer.py

[ROCm][CI] Fix moe layer tests for fp8 dtype compatibility (#56106)

2026-09-09 17:48:14 +00:00

test_moe_permute_unpermute.py

[MoE] Generalize masked activation for padded layouts (#51217)

2026-09-01 11:34:13 -04:00

test_moe_weight_loading_padded.py

[Bugfix][MoE] Fix fused block-scale orientation (#50727)

2026-08-11 16:05:38 -04:00

test_moe.py

[Quant][Kernel] Remove GPTQ Group/Dynamic Activation Ordering (#54809)

2026-09-08 12:45:12 -07:00

test_mxfp4_moe.py

Fix the E8M0 scale computation in the MXFP4 (W4A4) MOE CUTLASS kernel (#43557)

2026-06-15 06:04:54 -07:00

test_mxfp8_aiter_backend_selection.py

[CI] Fix MXFP8 MOE backend selection tests on gfx942 (#50222)

2026-07-29 17:17:11 +08:00

test_nvfp4_moe.py

[MoE] Remove inplace fused experts mechanism (#43727)

2026-05-27 20:00:19 -07:00

test_ocp_mx_moe.py

[ROCm][MoE] Split AITER CK and Triton MXFP4 W4A16 into separate backends (#50622)

2026-08-31 23:14:14 -05:00

test_profile_modular_kernel.py

[Bugfix] Fix broken profile_modular_kernel.py (#43300)

2026-06-11 12:17:23 +01:00

test_rocm_aiter_moe.py

[Bugfix][ROCm][MoE] Update AITER MXFP4 W4A16 tests to the renamed expert_mask (#51980)

2026-08-12 17:45:45 +00:00

test_rocm_aiter_topk.py

[MoE] Move various experts classes to fused_moe/experts/ (#41979)

2026-05-11 07:54:33 +08:00

test_routed_experts_capture_monolithic.py

[Bugfix] Gate sm_100-only kernel tests on the capability family, not >= (#54306)

2026-09-01 11:29:15 -04:00

test_routing_simulator.py

[MoE Refactor] Rename FusedMoE to FusedMoEFactory (#44941)

2026-07-31 11:28:18 +01:00

test_routing.py

Enable gfx1250 ROCm architecture (#46516)

2026-07-31 01:41:12 +00:00

test_shared_fused_moe_routed_transform.py

[MoE Refactor] Rename FusedMoE to FusedMoEFactory (#44941)

2026-07-31 11:28:18 +01:00

test_silu_mul_fp8_quant_deep_gemm.py

[ROCm] Add UE8M0 scale packing for Triton silu_mul_quant (#37835)

2026-08-19 10:20:05 -05:00

test_silu_mul_per_token_group_quant_fp8_colmajor.py

[XPU][Tests] Make tests device-agnostic (#51968)

2026-08-20 02:25:41 -07:00

test_situ_mul_fp8_quant.py

[MoE] Generalize masked activation for padded layouts (#51217)

2026-09-01 11:34:13 -04:00

test_topk_softplus_sqrt.py

[New model][Multimodal] Add DeepSeek-V4-Flash-Vision-Exp support (#54566)

2026-09-02 11:09:09 -07:00

test_triton_moe_no_act_mul.py

[XPU][Tests] Make tests device-agnostic (#51968)

2026-08-20 02:25:41 -07:00

test_triton_moe_ptpc_fp8.py

[CI][Bugfix] Fix test isolation in block_int8/ptpc_fp8 MoE kernel tests (#49609)

2026-07-23 12:26:02 -05:00

test_trtllm_bf16_moe.py

[Bugfix] Gate sm_100-only kernel tests on the capability family, not >= (#54306)

2026-09-01 11:29:15 -04:00

test_trtllm_nvfp4_moe.py

[Bugfix] Gate sm_100-only kernel tests on the capability family, not >= (#54306)

2026-09-01 11:29:15 -04:00

test_unquantized_backend_selection.py

[CPU] Migrate unquantized MoE to the modular-kernel experts structure (#50133)

2026-08-03 14:55:04 +08:00

test_zen_cpu_fused_moe.py

[Model] Add DeepSeek-V4 CPU backend (#55355)

2026-09-10 14:14:48 +08:00

test_zen_cpu_int8_moe.py

[Model] Add DeepSeek-V4 CPU backend (#55355)

2026-09-10 14:14:48 +08:00

test_zero_expert_moe.py

[MoE Refactor] Rename FusedMoE to FusedMoEFactory (#44941)

2026-07-31 11:28:18 +01:00

utils.py

[ROCm][CI] Fix moe layer tests for fp8 dtype compatibility (#56106)

2026-09-09 17:48:14 +00:00