Artem Perevedentsev 333529deae [EPLB] Fix replica selection bias in fused_moe router (#40810) ...

Signed-off-by: Artem Perevedentsev <aperevedents@nvidia.com>

2026-04-24 22:06:41 +00:00

..

configs

perf(moe): add tuned fused_moe config for RTX PRO 6000 Blackwell Server Edition (#39183)

2026-04-10 11:32:42 -06:00

experts

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

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

oracle

fused_moe: treat NIXL EP as batched experts (#40412)

2026-04-24 08:05:31 -05:00

prepare_finalize

[MoE] Move remaining PrepareAndFinalize to prepare finalize folder (#39009)

2026-04-23 20:00:53 -04:00

router

[EPLB] Fix replica selection bias in fused_moe router (#40810)

2026-04-24 22:06:41 +00:00

runner

[Bugfix][MoE] Unpad routed output before shared expert add [Fixes #35949] (#40794)

2026-04-24 11:53:23 +00:00

__init__.py

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

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

activation.py

[Refactor] Replace activation: str with MoEActivation enum (#33843)

2026-02-11 17:29:32 -08:00

all2all_utils.py

[MoE] Move remaining PrepareAndFinalize to prepare finalize folder (#39009)

2026-04-23 20:00:53 -04:00

config.py

fused_moe: treat NIXL EP as batched experts (#40412)

2026-04-24 08:05:31 -05:00

cpu_fused_moe.py

[CPU] Support gelu act in cpu_fused_moe (#38770)

2026-04-02 14:14:32 +08:00

deep_gemm_utils.py

[Bugfix] ep_scatter kernel store-load race condition (#34991)

2026-03-13 01:07:59 +00:00

fallback.py

[Feature]: Remove Chunking From FusedMoE (#34086)

2026-03-12 14:24:38 -04:00

flashinfer_cutlass_moe.py

Update flashinfer to 0.6.8 (#39959)

2026-04-20 10:37:23 -07:00

fused_batched_moe.py

[MoE] Move remaining PrepareAndFinalize to prepare finalize folder (#39009)

2026-04-23 20:00:53 -04:00

fused_humming_moe.py

[Quantization] add humming quantization kernel (#34556)

2026-04-24 09:29:44 -04:00

fused_marlin_moe.py

[MoE refactor] refactor GPTQMarlinMoEMethod with MK (#37990)

2026-04-23 05:21:47 +00:00

fused_moe_method_base.py

[MoE Refactor] Make SharedExperts class for use with DefaultMoERunner (#35153)

2026-04-01 09:44:08 -04:00

fused_moe_modular_method.py

[MoE Refactor] Make SharedExperts class for use with DefaultMoERunner (#35153)

2026-04-01 09:44:08 -04:00

fused_moe.py

[Refactor] Clean up log once scope="local" (#40540)

2026-04-22 16:42:43 -04:00

layer.py

[Quantization] add humming quantization kernel (#34556)

2026-04-24 09:29:44 -04:00

modular_kernel.py

[MoE Refactor] Move the shared/fused expert output sum into MoERunnerBase (#35949)

2026-04-20 12:28:46 -04:00

moe_align_block_size.py

[Kernel][MoE] optimize moe_align_block_size (#29642)

2025-12-07 01:58:47 -08:00

moe_fused_mul_sum.py

[Quantization] add humming quantization kernel (#34556)

2026-04-24 09:29:44 -04:00

moe_permute_unpermute.py

[Refactor] Remove align block size logic in moe_permute (#33449)

2026-02-06 10:57:06 -08:00

rocm_aiter_fused_moe.py

[ROCm] Cast score correction bias tensor during model construction for DeepSeek/Kimi-K2 (#39999)

2026-04-24 09:02:12 +09:00

routed_experts_capturer.py

fix(moe): fix RoutedExpertsCapturer assertion failure with DP>1 and MK path (#37879)

2026-04-12 10:28:17 -04:00

topk_weight_and_reduce.py

[MoE Refactor] Create MK for TRTLLM Kernels (#32564)

2026-03-03 10:39:50 -08:00

triton_cutlass_moe.py

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

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

triton_deep_gemm_moe.py

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

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

unquantized_fused_moe_method.py

[XPU] Quick fix for TritonMLA to remove cuda hardcode (#39088)

2026-04-08 00:17:58 +08:00

utils.py

[NVFP4] NVFP4 MOE emulation fallback for H100/MI300/MI350, standardize TritonExperts usage for OCP MX emulation (#35737)

2026-04-22 08:58:54 -07:00