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

[torch.compile] Unwrap fused_marlin_moe custom op (#26739)

Browse Source 
Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
This commit is contained in:
Varun Sundar Rabindranath
2025-10-13 22:22:16 -04:00
committed by GitHub
parent 8a0af6a561
commit 8ae169286f
10 changed files with 22 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/layers/quantization/gptq_marlin.py
View File

@@ -15,6 +15,7 @@ from vllm.model_executor.layers.fused_moe.config import (
FusedMoEConfig,
FusedMoEQuantConfig,
)
from vllm.model_executor.layers.fused_moe.fused_marlin_moe import fused_marlin_moe
from vllm.model_executor.layers.fused_moe.layer import (
FusedMoE,
FusedMoEMethodBase,
@@ -765,7 +766,7 @@ class GPTQMarlinMoEMethod(FusedMoEMethodBase):
indices_type=self.topk_indices_dtype,
)
return torch.ops.vllm.fused_marlin_moe(
return fused_marlin_moe(
x,
layer.w13_qweight,
layer.w2_qweight,