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 14, 05:30 UTC
[Perf] Dont create unnecessary pooling params (#22876)
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Lucas Wilkinson committed
Aug 14, 2025 at 08:28 UTC
829b9a62d0a89872883397ae4b5184048836589f
1 file changed
+3
-3
vllm/v1/worker/gpu_model_runner.py
+3
-3
@@ -341,13 +341,13 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
341
model_kwargs = dict[str, Any]()
342
num_reqs = self.input_batch.num_reqs
343
344
- pooling_params = self.input_batch.pooling_metadata.pooling_params
345
-
346
- num_pooling_reqs = len(pooling_params)
344
+ num_pooling_reqs = len(self.input_batch.pooling_params)
345
346
if num_pooling_reqs == 0:
347
return model_kwargs
348
349
+ pooling_params = self.input_batch.pooling_metadata.pooling_params
350
+
351
assert num_pooling_reqs == num_reqs
352
353
token_type_id_requests = dict[int, Any]()