Simon Mo
Simon Mo
> Any thoughts on testing? Good thinking, alternatively you can try trigger a release test pointing to a non-existent host like `wheel_url = "localhost:1234..."` which will for sure throw
@ray-project/ray-docs can you stamp this simple change to add a new monkey patched object in conf.py?
pinging @rkooo567 for review
Maybe we can consider an internal/experimental API? cc @iycheng as well.
@mvashishtha thanks for the ping. This is partially the expected behavior because log_to_driver is about your (user side) worker's print and logs. The unhandled error is propagated through the Ray...
> Job in Ray core doesn't support timeout yet. This is currently planned for 2.1 release. https://github.com/ray-project/ray/issues/22736
Pytest fixture helps modularize setup and tear down code. Generally considered more powerful. So most of the Ray codebase uses it except autoscaler and tune. Pytest is compatible with unittest...
If the "local directory" means a directory in the docker image, then it should work!
^my bad, didn't read the font cache part in readme, recompiling.
sorry i should have clarified the comment about namespace the exposed torch binding should be not be in cpp namespace https://github.com/vllm-project/vllm/blob/705578ae14b648782a8a321dd0903c163bd77375/csrc/quantization/awq/gemm_kernels.cu#L389-L394 all other helpers should include in `namespace vllm` ......