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 16, 07:00 UTC
[Bugfix][CPU] Fix CPU embedding runner with tensor parallel (#10394)
Signed-off-by: Isotr0py <2037008807@qq.com>
Isotr0py committed
Nov 17, 2024 at 15:12 UTC
cf349c4a97adb36354bdc2b14448ea55279d1575
1 file changed
+4
vllm/worker/cpu_embedding_model_runner.py
+4
@@ -66,6 +66,10 @@ class CPUEmbeddingModelRunner(
66
67
hidden_states = model_executable(**execute_model_kwargs)
68
69
+ # Only perform pooling in the driver worker.
70
+ if not self.is_driver_worker:
71
+ return []
72
+
73
return [
74
self.model.pooler(hidden_states=hidden_states,
75
pooling_metadata=model_input.pooling_metadata)