xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-09 21:57:41 +08:00
Code Issues Packages Projects Releases Wiki Activity

clean useless func comment

Browse Source 
Signed-off-by: zhuhaoran <zhuhaoran.zhr@alibaba-inc.com>
This commit is contained in:
zhuhaoran 2025-12-13 00:01:56 +08:00
parent 33c63f263d
commit 4672c4d035
2 changed files with 5 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/v1/worker/gpu_input_batch.py
View File

@ -971,10 +971,9 @@ class InputBatch:
num_draft_tokens: list[int] | None = None, num_draft_tokens: list[int] | None = None,
) -> None: ) -> None:
""" """
In async scheduling case, update spec_token_ids in sampling metadata In async scheduling case, update spec_token_ids in sampling metadata with
with real draft token ids from prior step. real draft token ids from prior step. This is called right before they are
This is called right before they are needed by the rejection sampler needed by the rejection sampler for penalty/bad_words computation.
for penalty/bad_words computation.
""" """
if draft_token_ids_cpu is None or self.prev_req_id_to_index is None: if draft_token_ids_cpu is None or self.prev_req_id_to_index is None:
return return

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

@ -3403,12 +3403,7 @@ class GPUModelRunner(
def _copy_draft_token_ids( def _copy_draft_token_ids(
self, draft_token_ids: torch.Tensor, num_reqs: int self, draft_token_ids: torch.Tensor, num_reqs: int
) -> None: ) -> None:
"""Copy draft token ids to CPU asynchronously. """Copy draft token ids to CPU asynchronously."""
This is used for async scheduling with spec decode + penalty/bad_words.
The draft_token_ids will be used in the next step to update
input_batch.spec_token_ids for correct penalty/bad_words computation.
"""
if self.draft_token_ids_copy_event is None or not isinstance( if self.draft_token_ids_copy_event is None or not isinstance(
draft_token_ids, torch.Tensor draft_token_ids, torch.Tensor
): ):
@ -3426,12 +3421,7 @@ class GPUModelRunner(
self.draft_token_ids_copy_event.record() self.draft_token_ids_copy_event.record()
def _get_draft_token_ids_cpu(self) -> list[list[int]] | None: def _get_draft_token_ids_cpu(self) -> list[list[int]] | None:
"""Get previously copied draft token ids from CPU. """Get previously copied draft token ids from CPU."""
Called at the beginning of the next step to update spec_token_ids
for async scheduling with spec decode + penalty/bad_words.
Returns None if no draft tokens were copied in previous step.
"""
if isinstance(self._draft_token_ids, list): if isinstance(self._draft_token_ids, list):
return self._draft_token_ids return self._draft_token_ids