[Bugfix] Pass vllm_config to kv_connector_no_forward in NPUModelRunner (#4970)

jianzs · web-flow · commit 45889a618599 · 2025-12-12T22:36:23.000+08:00
### What this PR does / why we need it? The newest version crashes in PD separation scenarios because the function is missing the `vllm_config` parameter. - vLLM version: v0.12.0 - vLLM main: vllm-project/vllm@ad32e3e Signed-off-by: Jade Zheng <zheng.shoujian@outlook.com>
diff --git a/vllm_ascend/worker/model_runner_v1.py b/vllm_ascend/worker/model_runner_v1.py
@@ -1444,7 +1444,8 @@ def execute_model(
                     )
                     # Return empty ModelRunnerOuptut if there's no work to do.
                     return EMPTY_MODEL_RUNNER_OUTPUT
-                return self.kv_connector_no_forward(scheduler_output)
+                return self.kv_connector_no_forward(scheduler_output,
+                                                    self.vllm_config)
 
             if self.dynamic_eplb:
                 self.eplb_updator.forward_before()

Original file line number	Diff line number	Diff line change
`@@ -1444,7 +1444,8 @@ def execute_model(`
`1444`	`1444`	`)`
`1445`	`1445`	`# Return empty ModelRunnerOuptut if there's no work to do.`
`1446`	`1446`	`return EMPTY_MODEL_RUNNER_OUTPUT`
`1447`		`- return self.kv_connector_no_forward(scheduler_output)`
	`1447`	`+ return self.kv_connector_no_forward(scheduler_output,`
	`1448`	`+ self.vllm_config)`
`1448`	`1449`
`1449`	`1450`	`if self.dynamic_eplb:`
`1450`	`1451`	`self.eplb_updator.forward_before()`