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

[Model] Add LongCat-Flash (#23991)

Browse Source 
Signed-off-by: yangxurui <yangxurui@meituan.com>
Co-authored-by: yangxurui <yangxurui@meituan.com>
This commit is contained in:
XuruiYang
2025-09-25 12:53:40 +08:00
committed by GitHub
parent 90b139cfff
commit 845adb3ec6
31 changed files with 1357 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/model_executor/layers/quantization/mxfp4.py
View File

@@ -718,7 +718,7 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
assert isinstance(self.fused_experts, mk.FusedMoEModularKernel)
topk_weights, topk_ids = FusedMoE.select_experts(
topk_weights, topk_ids, _ = FusedMoE.select_experts(
hidden_states=x,
router_logits=router_logits,
use_grouped_topk=use_grouped_topk,
@@ -783,7 +783,7 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
raise NotImplementedError("EPLB is not supported for mxfp4")
if self.mxfp4_backend == Mxfp4Backend.MARLIN:
topk_weights, topk_ids = FusedMoE.select_experts(
topk_weights, topk_ids, _ = FusedMoE.select_experts(
hidden_states=x,
router_logits=router_logits,
use_grouped_topk=use_grouped_topk,
@@ -894,7 +894,7 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
or self.mxfp4_backend == Mxfp4Backend.SM90_FI_MXFP4_BF16):
from vllm.utils.flashinfer import flashinfer_cutlass_fused_moe
topk_weights, topk_ids = FusedMoE.select_experts(
topk_weights, topk_ids, _ = FusedMoE.select_experts(
hidden_states=x,
router_logits=router_logits,
use_grouped_topk=use_grouped_topk,