From 7feb74590b63206fa078d28dfdaead74ce412645 Mon Sep 17 00:00:00 2001 From: Angazenn <92204292+Angazenn@users.noreply.github.com> Date: Fri, 16 Jan 2026 23:29:35 +0800 Subject: [PATCH] Revert "[bugfix]limit graph replay sync (#5761)" (#5965) ### What this PR does / why we need it? reverts #5761 to fix accuracy issues when using piecewise graph mode. ### Does this PR introduce _any_ user-facing change? ### How was this patch tested? - vLLM version: v0.13.0 - vLLM main: https://github.com/vllm-project/vllm/commit/2c24bc6996cb165fce92f780b388a5e39b3f4060 Signed-off-by: Angazenn --- vllm_ascend/compilation/acl_graph.py | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/vllm_ascend/compilation/acl_graph.py b/vllm_ascend/compilation/acl_graph.py index c81d2d2f..28eafed5 100644 --- a/vllm_ascend/compilation/acl_graph.py +++ b/vllm_ascend/compilation/acl_graph.py @@ -186,13 +186,12 @@ class ACLGraphWrapper: ) logger.info_once("Replaying aclgraph") - # In async scheduling or multi-threaded (MT) scenarios when graph mode is FULL, it is possible that + # In async scheduling or multi-threaded (MT) scenarios, it is possible that # the CPU's record event (from update_attn_params) for the iteration i completes # before the grph replay of iteration i-1. # To ensure proper ordering, we must call synchronize here before replaying, # so that update_attn_params only executes after the previous graph replay has fully completed. - if self.runtime_mode == CUDAGraphMode.FULL: - torch.npu.synchronize() + torch.npu.synchronize() entry.aclgraph.replay() return entry.output