Skip to content

Commit 5605422

Browse files
committed
add logger.info to indicate the completion of kv transfer
1 parent 35c22a2 commit 5605422

File tree

1 file changed

+2
-0
lines changed

1 file changed

+2
-0
lines changed

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -821,6 +821,7 @@ def get_finished(self, scheduler_output) -> tuple[set[str], set[str]]:
821821
requests = self._transfering_req_meta[req]
822822
for request_id, reqmeta in requests.items():
823823
assert len(reqmeta.local_block_ids) == len(reqmeta.remote_block_ids)
824+
logger.info(f'get_finished in NIXL: {request_id=}|{reqmeta=}')
824825
for i, local_block_id in enumerate(reqmeta.local_block_ids):
825826
#print(f'buke get_finished: {local_block_id=}||{len(self.kv_caches_cpu)=}')
826827
for idx, (layer, kv_layer) in enumerate(self.kv_caches_cpu.items()):
@@ -833,6 +834,7 @@ def get_finished(self, scheduler_output) -> tuple[set[str], set[str]]:
833834
self.kv_caches_hpu[layer][1][start:end].copy_(v[start:end], non_blocking = False)
834835
k00,v00 = self.kv_caches_hpu['model.layers.0.self_attn.attn']
835836
del self._transfering_req_meta[req]
837+
836838
logger.debug(f'buke: get_finished hpu: {k00.shape=}|{k00.sum(dim=[1,2])[100:400]=}')
837839
logger.debug(
838840
"Rank %s, get_finished: %s requests done sending "

0 commit comments

Comments
 (0)