vllm/worker/tpu_model_runner.py · ce6bf3a2cff4860c5661cac2280e0a28bedb6440 · OpenDAS / vllm_cscc · GitLab

Find file Blame History Permalink

[torch.compile] avoid Dynamo guard evaluation overhead (#7898) · ce6bf3a2
youkaichao authored Aug 28, 2024
```
Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
```
ce6bf3a2

tpu_model_runner.py 32 KB