biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[GPTOSS][DP/EP][Marlin] Enable GPTOSS DP/EP using Marlin kernels (#25488)

Browse Source 
Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Varun Sundar Rabindranath
2025-10-03 20:13:13 -04:00
committed by GitHub
parent 767cbb011d
commit 7ef40bb983
9 changed files with 264 additions and 101 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
docs/design/moe_kernel_features.md
View File

@@ -93,6 +93,8 @@ To be used with a particular `FusedMoEPrepareAndFinalize` sub-class, MoE kernels
| gpt oss triton | standard | N/A | N/A | <sup>5</sup> | Y | Y | [`triton_kernel_fused_experts`][vllm.model_executor.layers.fused_moe.gpt_oss_triton_kernels_moe.triton_kernel_fused_experts],</br>[`OAITritonExperts`][vllm.model_executor.layers.fused_moe.gpt_oss_triton_kernels_moe.OAITritonExperts] |
| deep gemm+triton<sup>2</sup> | standard,</br>batched | all<sup>1</sup> | G(128),A,T | silu, gelu | <sup>6</sup> | Y | [`TritonOrDeepGemmExperts`][vllm.model_executor.layers.fused_moe.triton_deep_gemm_moe.TritonOrDeepGemmExperts],</br>[`BatchedTritonOrDeepGemmExperts`][vllm.model_executor.layers.fused_moe.batched_triton_or_deep_gemm_moe.BatchedTritonOrDeepGemmExperts] |
| marlin | standard | <sup>3</sup> | <sup>3</sup> | silu,</br>swigluoai | Y | N | [`fused_marlin_moe`][vllm.model_executor.layers.fused_moe.fused_marlin_moe.fused_marlin_moe] |
| marlin experts | standard | N/A | N/A | silu,</br>swigluoai | Y | Y | [`MarlinExperts`][vllm.model_executor.layers.fused_moe.fused_marlin_moe.MarlinExperts] |
| trtllm | standard | mxfp4,</br>nvfp4 | G(16),G(32) | <sup>5</sup> | N | Y | [`TrtLlmGenExperts`][vllm.model_executor.layers.fused_moe.trtllm_moe.TrtLlmGenExperts] |
| pallas | standard | N/A | N/A | silu | N | N | [`fused_moe`][vllm.model_executor.layers.fused_moe.moe_pallas.fused_moe] |
| iterative | standard | N/A | N/A | silu | N | N | [`fused_moe`][vllm.model_executor.layers.fused_moe.moe_torch_iterative.fused_moe] |
@@ -114,6 +116,6 @@ The following table shows "families" of modular kernels that are intended to wor
| backend | `FusedMoEPrepareAndFinalize` subclasses | `FusedMoEPermuteExpertsUnpermute` subclasses |
|----------------------------------|------------------------------------------------------------|----------------------------------------------------------------------------------------------------------------------------|
| deepep_high_throughput,</br>pplx | `DeepEPHTPrepareAndFinalize`,</br>`PplxPrepareAndFinalize` | `BatchedDeepGemmExperts`,</br>`BatchedTritonExperts`,</br>`BatchedTritonOrDeepGemmExperts`,</br>`CutlassBatchedExpertsFp8` |
| deepep_low_latency | `DeepEPLLPrepareAndFinalize` | `DeepGemmExperts`,</br>`TritonExperts`,</br>`TritonOrDeepGemmExperts`,</br>`CutlassExpertsFp8` |
| deepep_high_throughput | `DeepEPHTPrepareAndFinalize` | `DeepGemmExperts`,</br>`TritonExperts`,</br>`TritonOrDeepGemmExperts`,</br>`CutlassExpertsFp8`, </br>`MarlinExperts` |
| deepep_low_latency,</br>pplx | `DeepEPLLPrepareAndFinalize`,</br>`PplxPrepareAndFinalize` | `BatchedDeepGemmExperts`,</br>`BatchedTritonExperts`,</br>`BatchedTritonOrDeepGemmExperts`,</br>`CutlassBatchedExpertsFp8`|
| flashinfer | `FlashInferCutlassMoEPrepareAndFinalize` | `FlashInferExperts` |