xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-24 23:25:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Misc] Avoid redundant copy for encoder-only models (#24012)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon 2025-08-31 21:02:43 -07:00 committed by GitHub
parent 183a70967a
commit 8c742a66d1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/v1/worker/gpu_model_runner.py
View File

@ -827,13 +827,13 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
blk_table_tensor = torch.zeros(
(num_reqs, 1),
dtype=torch.int32,
pin_memory=self.pin_memory,
device="cpu").to(self.device, non_blocking=True)
slot_mapping = torch.zeros((total_num_scheduled_tokens, ),
dtype=torch.int32,
pin_memory=self.pin_memory,
device="cpu").to(self.device,
non_blocking=True)
device=self.device,
)
slot_mapping = torch.zeros(
(total_num_scheduled_tokens, ),
dtype=torch.int64,
device=self.device,
)
num_common_prefix_blocks = 0
else:
blk_table = self.input_batch.block_table[kv_cache_group_id]