xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-09 15:25:48 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Core] Simplify setting new_token_ids in CachedRequestData (#26388)

Browse Source 
Signed-off-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Nick Hill 2025-10-07 20:32:37 -07:00 committed by GitHub
parent 046118b938
commit 067da2d1df
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/v1/core/sched/scheduler.py
View File

@ -712,7 +712,6 @@ class Scheduler(SchedulerInterface):
num_computed_tokens: list[int] = []
num_output_tokens: list[int] = []
use_connector = self.connector is not None
for req in itertools.chain(running_reqs, resumed_reqs):
req_id = req.request_id
req_ids.append(req_id)
@ -729,16 +728,11 @@ class Scheduler(SchedulerInterface):
req.num_computed_tokens : req.num_computed_tokens + num_tokens
]
new_token_ids.append(token_ids)
elif use_connector:
# When using a KVConnector, we add a placeholder to avoid index
# out of bounds errors. TODO: Remove this once the KVConnector
# is updated to handle token IDs properly.
new_token_ids.append([])
new_block_ids.append(
req_to_new_blocks[req_id].get_block_ids(allow_none=True)
)
num_computed_tokens.append(req.num_computed_tokens)
num_output_tokens.append(len(req.output_token_ids))
num_output_tokens.append(req.num_output_tokens)
# Because resumed_reqs is usually empty, it is more efficient to do
# in-place appending so that we don't need to allocate a new list.
resumed_from_preemption = [False] * len(running_reqs)