EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Bugfix] fix mtp and eagle aclgraph bug (#4710)

Browse Source 
### What this PR does / why we need it?
fix mtp and eagle aclgraph bug

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

Signed-off-by: GDzhu01 <809721801@qq.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
Zhu Yi Lin
2025-12-06 11:22:57 +08:00
committed by GitHub
parent 74033999ed
commit f067623afd
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm_ascend/spec_decode/eagle_proposer.py
View File

@@ -51,7 +51,7 @@ class EagleProposer(Proposer):
not self.vllm_config.model_config.enforce_eager)
self.cudagraph_batch_sizes = list(
reversed(
sorted(
self.vllm_config.compilation_config.cudagraph_capture_sizes))
# persistent buffers for cuda graph

2
vllm_ascend/spec_decode/mtp_proposer.py
View File

@@ -108,7 +108,7 @@ class MtpProposer(Proposer):
self.use_aclgraph = self.runner._use_aclgraph()
self.cudagraph_batch_sizes = (list(
reversed(
sorted(
self.vllm_config.compilation_config.cudagraph_capture_sizes))
if self.use_aclgraph else [])