Commits

Branch selector

 main

User selector

bnellnm

All time

Commit History

Commits on Apr 23, 2026

• [Bugfix] Fix DeepSeek V2-Lite Accuracy drop (#40673)

  Show description for 4a6dd1c
  bnellnm

  authored
  4a6dd1c

  Copy full SHA for 4a6dd1c

• [MoE Refactor] Rename FusedMoE.make_expert_params_mapping to fused_moe_make_expert_params_mapping (#40671)

  Show description for 1c2c1eb
  bnellnm

  authored
  1c2c1eb

  Copy full SHA for 1c2c1eb

Commits on Apr 22, 2026

• [MoE Refactor] Combine MoERunnerBase + DefaultMoERunner (#40560)

  Show description for 809d83c
  bnellnm

  authored
  809d83c

  Copy full SHA for 809d83c

Commits on Apr 21, 2026

• [MoE Refactor] Add more MoE layer tests (#39349)

  Show description for 9db4650
  

  bnellnm

  and

  robertgshaw2-redhat

  authored
  9db4650

  Copy full SHA for 9db4650

• [MoE Refactor] Remove SharedFusedMoE class (#35782)

  Show description for 5e584ce
  bnellnm

  authored
  5e584ce

  Copy full SHA for 5e584ce

Commits on Apr 20, 2026

• [Bugfix] Replace code that disabled shared expert overlap (#39222)

  Show description for 6867bcd
  

  bnellnm

  and

  robertgshaw2-redhat

  authored
  6867bcd

  Copy full SHA for 6867bcd

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

  Show description for 726efe1
  

  bnellnm

  and

  robertgshaw2-redhat

  authored
  726efe1

  Copy full SHA for 726efe1

Commits on Apr 16, 2026

• [Bugfix] Temporarily disable B200 fp4 MoE layer tests (#40057)

  Show description for 79e799e
  bnellnm

  authored
  79e799e

  Copy full SHA for 79e799e

Commits on Apr 14, 2026

• [MoE Refactor] Refactor ZeroExpertFusedMoE into new framework (#35549)

  Show description for 19ec9a0
  bnellnm

  authored
  19ec9a0

  Copy full SHA for 19ec9a0

• [MoE Refactor] Remove MoE DP chunking (#39107)

  Show description for e1e318a
  

  bnellnm

  and

  robertgshaw2-redhat

  authored
  e1e318a

  Copy full SHA for e1e318a

Commits on Apr 7, 2026

• [MoE Refactor] Split up compressed_tensors_moe.py (#38960)

  Show description for b2b2c52
  bnellnm

  authored
  b2b2c52

  Copy full SHA for b2b2c52

Commits on Apr 6, 2026

• [MoE Refactor][Test] FusedMoE layer test (#24675)

  Show description for f014824
  

  bnellnm

  and

  robertgshaw2-redhat

  authored
  f014824

  Copy full SHA for f014824

• [MoE Refactor] Split of DefaultMoERunner class (#35326)

  Show description for 93bada4
  

  bnellnm

  and

  robertgshaw2-redhat

  authored
  93bada4

  Copy full SHA for 93bada4

Commits on Apr 1, 2026

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

  Show description for 7cf56a5
  bnellnm

  authored
  7cf56a5

  Copy full SHA for 7cf56a5

Commits on Mar 19, 2026

• [MoE Refactor] Rename "naive" all2all backend (#36294)

  Show description for 91be5f9
  bnellnm

  authored
  91be5f9

  Copy full SHA for 91be5f9

• [MoE Refactor] DefaultMoERunner simplifcation (#33049)

  Show description for 9279c59
  bnellnm

  authored
  9279c59

  Copy full SHA for 9279c59

Commits on Mar 3, 2026

• [Bugfix] Fix misnamed parameter in compressed_tensors_moe.py (#35813)

  Show description for a9b8b13
  

  bnellnm

  and

  robertgshaw2-redhat

  authored
  a9b8b13

  Copy full SHA for a9b8b13

Commits on Feb 16, 2026

• [Bugfix] Add method to swap quant_method on FusedMoE to fix LoRA issues (#34453)

  Show description for 5bff999
  bnellnm

  authored
  5bff999

  Copy full SHA for 5bff999

Commits on Feb 13, 2026

• [Bugfix] Remove assert that's no longer valid (#34443)

  Show description for 04ea31b
  bnellnm

  authored
  04ea31b

  Copy full SHA for 04ea31b

Commits on Feb 11, 2026

• [Bugfix] Fix some issues with MoERunner PR #32344 (#34371)

  Show description for 31d992d
  bnellnm

  authored
  31d992d

  Copy full SHA for 31d992d

• [MoE Refactor] Introduce MoERunner abstraction and move execution logic from FusedMoE to DefaultMoERunner (#32344)

  Show description for d1481ba
  bnellnm

  authored
  d1481ba

  Copy full SHA for d1481ba

Commits on Feb 5, 2026

• [Moe Refactor] Make Inplace Flag for FusedMoEModularKernel part of the constructor (#33375)

  Show description for a57c822
  

  bnellnm

  and

  robertgshaw2-redhat

  authored
  a57c822

  Copy full SHA for a57c822

Commits on Jan 22, 2026

• [MoE Refactor] Move `select_experts` from `FusedMoEQuantMethod` -> `FusedMoE` (#31996)

  Show description for dc917cc
  bnellnm

  authored
  dc917cc

  Copy full SHA for dc917cc

Commits on Jan 18, 2026

• [MoE Refactor] Separate Router into OO Classes (#30623)

  Show description for 327a02d
  bnellnm

  authored
  327a02d

  Copy full SHA for 327a02d

Commits on Jan 8, 2026

• [Misc][Refactor] Add FusedMoERouter object (#30519)

  Show description for e74698c
  bnellnm

  authored
  e74698c

  Copy full SHA for e74698c

Commits on Dec 9, 2025

• [MoE][Refactor] Remove most arguments to FusedMoEMethodBase.apply (#29066)

  Show description for 00e5cbb
  

  4 peopleauthored
  00e5cbb

  Copy full SHA for 00e5cbb

Commits on Dec 3, 2025

• [Kernels] Remove BatchedTritonOrDeepGemmExperts and default fallback to Triton (#29929)

  Show description for 2902c34
  

  bnellnm

  and

  tlrmchlsmth

  authored
  2902c34

  Copy full SHA for 2902c34

Commits on Nov 24, 2025

• [MoE][Refactor] Make select_experts a non-static method (#29067)

  Show description for 8f06614
  bnellnm

  authored
  8f06614

  Copy full SHA for 8f06614

Commits on Nov 11, 2025

• [Kernels] Split up fused_moe/layer.py, isolate more modular kernel code (#28064)

  bnellnm

  authored
  a1448b4

  Copy full SHA for a1448b4

Commits on Nov 4, 2025

• [Kernels] Isolate modular kernel code from FusedMoEMethodBase subclasses. (#27123)

  bnellnm

  authored
  938772a

  Copy full SHA for 938772a

Commits on Oct 29, 2025

• [Bugfix] Fix modular kernel tests (#27707)

  Show description for 1891cf6
  bnellnm

  authored
  1891cf6

  Copy full SHA for 1891cf6

Commits on Oct 13, 2025

• [Misc] cache result of disable_inplace (#26666)

  Show description for 60e419c
  bnellnm

  authored
  60e419c

  Copy full SHA for 60e419c

Commits on Oct 9, 2025

• [Bugfix] Disable moe inplace for torch >= 2.9 (#26497)

  Show description for a462331
  bnellnm

  authored
  a462331

  Copy full SHA for a462331

• [Model] Apply shared experts overlap optimization to all models with shared experts (#26145)

  Show description for 47e66c2
  bnellnm

  authored
  47e66c2

  Copy full SHA for 47e66c2

Commits on Oct 8, 2025

• [Kernels] Modular kernel refactor (#24812)

  Show description for da36461
  bnellnm

  authored
  da36461

  Copy full SHA for da36461