A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 12, 13:30 UTC
dummy run corner case (#29433)
Xieyang Xu committed
Nov 25, 2025 at 16:20 UTC
12866af748f3933b62891f02646526c64395b26f
1 file changed
+1
-1
vllm/v1/worker/gpu_model_runner.py
+1
-1
@@ -2789,7 +2789,7 @@ class GPUModelRunner(
2789
# returns True. before returning early here we call
2790
# dummy run to ensure coordinate_batch_across_dp
2791
# is called into to avoid out of sync issues.
2792
- self._dummy_run(1)
2792
+ self._dummy_run(self._get_num_input_tokens(1))
2793
if not has_kv_transfer_group():
2794
# Return empty ModelRunnerOutput if no work to do.
2795
return EMPTY_MODEL_RUNNER_OUTPUT