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, 02:00 UTC
[Misc]Minor Changes about Worker (#11555)
Signed-off-by: Chenguang Li <757486878@qq.com>
Chenguang Li committed
Jan 13, 2025 at 15:47 UTC
c3f05b09a040b9d13ad62914be3f7a84c535e417
2 files changed
-2
vllm/v1/worker/gpu_worker.py
-1
@@ -132,7 +132,6 @@ class Worker:
132
# Execute a forward pass with dummy inputs to profile the memory usage
133
# of the model.
134
self.model_runner.profile_run()
135
- torch.cuda.synchronize()
135
136
free_gpu_memory, _ = torch.cuda.mem_get_info()
137
# NOTE(woosuk): Here we assume that the other processes using the same
vllm/worker/worker.py
-1
@@ -200,7 +200,6 @@ class Worker(LocalOrDistributedWorkerBase):
200
weights_memory_in_bytes=self.model_runner.
201
model_memory_usage) as result:
202
self.model_runner.profile_run()
203
- torch.cuda.synchronize()
203
204
self._assert_memory_footprint_increased_during_profiling()
205