xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-05 02:16:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

[XPU] Enable custom routing functions in IPEX for Llama4 (#28004)

Browse Source 
Signed-off-by: frost-intel <frost.mitchell@intel.com>
This commit is contained in:
Frost Mitchell 2025-11-05 08:39:57 -05:00 committed by GitHub
parent 6ab183813c
commit 6e97eccf5d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/model_executor/layers/fused_moe/layer.py
View File

@ -915,7 +915,6 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
or logical_replica_count is not None or logical_replica_count is not None
): ):
raise NotImplementedError("Expert load balancing is not supported for XPU.") raise NotImplementedError("Expert load balancing is not supported for XPU.")
assert custom_routing_function is None
return layer.ipex_fusion( return layer.ipex_fusion(
x, x,
use_grouped_topk, use_grouped_topk,
@ -924,6 +923,7 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
renormalize, renormalize,
topk_group, topk_group,
num_expert_group, num_expert_group,
custom_routing_function=custom_routing_function,
) )
def forward_tpu( def forward_tpu(