[Feat]support sequence parallelism by pass for VL models (#5632)

2026-02-27 08:27:41 +08:00
parent ed175d6d92
commit 5def28dcd3
22 changed files with 460 additions and 101 deletions
--- a/vllm_ascend/compilation/graph_fusion_pass_manager.py
+++ b/vllm_ascend/compilation/graph_fusion_pass_manager.py
@@ -70,3 +70,8 @@ class GraphFusionPassManager:
            from .passes.allreduce_rmsnorm_fusion_pass import MatmulAllReduceAddRMSNormPass

            self.passes.append(MatmulAllReduceAddRMSNormPass(config))
+
+        if config.compilation_config.pass_config.enable_sp:
+            from .passes.sequence_parallelism import AscendSequenceParallelismPass
+
+            self.passes.append(AscendSequenceParallelismPass(config))