youngkingdom/vllm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[V1] Simplify GPUModelRunner._update_states check (#13265)

Browse Source
This commit is contained in:
Nick Hill
2025-02-14 04:33:43 -08:00
committed by GitHub
parent 4da1f667e9
commit 085b7b2d6c
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/v1/worker/gpu_model_runner.py
View File

@ -347,6 +347,8 @@ class GPUModelRunner(LoRAModelRunnerMixin):
self.input_batch.block_table.append_row(req_index, start_index,
req_data.new_block_ids)
batch_changed = len(removed_req_indices) > 0 or len(req_ids_to_add) > 0
# Add the new or resumed requests to the persistent batch.
# The smaller empty indices are filled first.
removed_req_indices = sorted(removed_req_indices, reverse=True)
@ -363,8 +365,8 @@ class GPUModelRunner(LoRAModelRunnerMixin):
# Condense the batched states if there are empty indices.
if removed_req_indices:
self.input_batch.condense(removed_req_indices)
return (len(unscheduled_req_ids) > 0 or len(req_ids_to_add) > 0
or len(scheduler_output.finished_req_ids) > 0)
return batch_changed
def _prepare_inputs(self, scheduler_output: "SchedulerOutput"):
total_num_scheduled_tokens = scheduler_output.total_num_scheduled_tokens