frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

cleanup

Browse Source 
Signed-off-by: Robert Shaw <robshaw@redhat.com>
This commit is contained in:
Robert Shaw
2025-07-07 00:30:51 +00:00
parent 7fbcbbfc45
commit c22a6cb1cc
1 changed files with 2 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
View File

@ -37,9 +37,6 @@ if TYPE_CHECKING:
Transfer = tuple[int, float] # (xfer_handle, start_time)
GET_META_MSG = b"get_meta_msg"
NIXL_MAX_DESCS = 1000
import os
USE_BATCHED = os.getenv("USE_BATCHED", "1") == "1"
logger = init_logger(__name__)
@ -1003,15 +1000,9 @@ class NixlConnectorWorker:
# Begin async xfer.
start = time.perf_counter()
if USE_BATCHED:
print("BATCHED!")
self.nixl_wrapper.transfer_batched(handles)
else:
print("NON BATCHED!")
for handle in handles:
self.nixl_wrapper.transfer(handle)
self.nixl_wrapper.transfer_batched(handles)
end = time.perf_counter()
logger.info("======== LAUNCH TIME: %s ========", end - start)
logger.info("========== TRANSFER BATCHED: %s ==========", end - start)
# Keep track of ongoing transfers.
remote_rank = self.tp_rank // tp_ratio