From f067623afd3f4ff94df35c2ec8779df3d60496b8 Mon Sep 17 00:00:00 2001 From: Zhu Yi Lin <116337067+GDzhu01@users.noreply.github.com> Date: Sat, 6 Dec 2025 11:22:57 +0800 Subject: [PATCH] [Bugfix] fix mtp and eagle aclgraph bug (#4710) ### What this PR does / why we need it? fix mtp and eagle aclgraph bug - vLLM version: v0.12.0 - vLLM main: https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9 Signed-off-by: GDzhu01 <809721801@qq.com> Co-authored-by: Mengqing Cao Co-authored-by: wangxiyuan --- vllm_ascend/spec_decode/eagle_proposer.py | 2 +- vllm_ascend/spec_decode/mtp_proposer.py | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/vllm_ascend/spec_decode/eagle_proposer.py b/vllm_ascend/spec_decode/eagle_proposer.py index 2dd1a7d6..6b47e6bf 100644 --- a/vllm_ascend/spec_decode/eagle_proposer.py +++ b/vllm_ascend/spec_decode/eagle_proposer.py @@ -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 diff --git a/vllm_ascend/spec_decode/mtp_proposer.py b/vllm_ascend/spec_decode/mtp_proposer.py index 89425c6a..4b7bfad9 100644 --- a/vllm_ascend/spec_decode/mtp_proposer.py +++ b/vllm_ascend/spec_decode/mtp_proposer.py @@ -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 [])