[Bugfix][PD]fix non-working disaggregated prefill (#2374)

### What this PR does / why we need it?

Mainline vLLM fixes its disaggregated prefill in
https://github.com/vllm-project/vllm/pull/22598 . But it is still not
working in vllm-ascend.
To be concrete, decoder instances crash before vllm's fix and hang after
vllm's fix in ascend devices.
This patch allows disaggregated prefill to work.

### Does this PR introduce _any_ user-facing change?
No.

### How was this patch tested?
Qwen3-0.6B 1P1D tp=1 dp=1


- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/0fe85087a950f3ca94d60293c865c0e6c05e6eff

---------

Signed-off-by: CaveNightingale <cavenightingale@foxmail.com>
This commit is contained in:
CaveNightingale authored and GitHub committed 2025-08-15 16:59:52 +08:00
1 parent 1b40665548
commit 2bb7e55022
1 file changed
+4 -3
+4 -3
View File
@@ -1636,7 +1636,7 @@ class NPUModelRunner(LoRAModelRunnerMixin):
finished_recving) = (self._process_reqs(scheduler_output,
intermediate_tensors))
kv_connector_output = None
if finished_sending is not None and finished_recving is not None:
if finished_sending is not None or finished_recving is not None:
kv_connector_output = KVConnectorOutput(
finished_sending=finished_sending,
finished_recving=finished_recving)
@@ -1838,8 +1838,9 @@ class NPUModelRunner(LoRAModelRunnerMixin):
return EMPTY_MODEL_RUNNER_OUTPUT
output = copy.copy(EMPTY_MODEL_RUNNER_OUTPUT)
output.finished_sending = finished_sending
output.finished_recving = finished_recving
output.kv_connector_output = KVConnectorOutput(
finished_sending=finished_sending,
finished_recving=finished_recving)
return output
@staticmethod