xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-03 15:24:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

lint: fix mypy error

Browse Source 
Signed-off-by: zhuhaoran <zhuhaoran.zhr@alibaba-inc.com>
This commit is contained in:
zhuhaoran 2025-12-12 10:56:06 +08:00
parent 0490418742
commit aec572f39d
1 changed files with 6 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
vllm/v1/worker/gpu_model_runner.py
View File

@ -591,7 +591,6 @@ class GPUModelRunner(
# with dedicated stream for overlapping and event for coordination. # with dedicated stream for overlapping and event for coordination.
self.valid_sampled_token_count_event: torch.Event | None = None self.valid_sampled_token_count_event: torch.Event | None = None
self.valid_sampled_token_count_copy_stream: torch.cuda.Stream | None = None self.valid_sampled_token_count_copy_stream: torch.cuda.Stream | None = None
self.valid_sampled_token_count_cpu: torch.Tensor | None = None
# Pre-allocated tensor for copying draft token ids to CPU, # Pre-allocated tensor for copying draft token ids to CPU,
# with dedicated stream for overlapping and event for coordination. # with dedicated stream for overlapping and event for coordination.
self.draft_token_ids_copy_event: torch.Event | None = None self.draft_token_ids_copy_event: torch.Event | None = None
@ -600,12 +599,6 @@ class GPUModelRunner(
if self.use_async_scheduling and self.num_spec_tokens: if self.use_async_scheduling and self.num_spec_tokens:
self.valid_sampled_token_count_event = torch.Event() self.valid_sampled_token_count_event = torch.Event()
self.valid_sampled_token_count_copy_stream = torch.cuda.Stream() self.valid_sampled_token_count_copy_stream = torch.cuda.Stream()
self.valid_sampled_token_count_cpu = torch.empty(
self.max_num_reqs,
dtype=torch.int64,
device="cpu",
pin_memory=self.pin_memory,
)
self.draft_token_ids_copy_event = torch.Event() self.draft_token_ids_copy_event = torch.Event()
self.draft_token_ids_copy_stream = torch.cuda.Stream() self.draft_token_ids_copy_stream = torch.cuda.Stream()
self.draft_token_ids_cpu = torch.empty( self.draft_token_ids_cpu = torch.empty(
@ -615,6 +608,12 @@ class GPUModelRunner(
pin_memory=self.pin_memory, pin_memory=self.pin_memory,
) )
self._prev_copy_draft_num_reqs: int = 0 self._prev_copy_draft_num_reqs: int = 0
self.valid_sampled_token_count_cpu = torch.empty(
self.max_num_reqs,
dtype=torch.int64,
device="cpu",
pin_memory=self.pin_memory,
)
# Ephemeral state transferred between execute_model() and sample_tokens(). # Ephemeral state transferred between execute_model() and sample_tokens().
self.execute_model_state: ExecuteModelState | None = None self.execute_model_state: ExecuteModelState | None = None