[AWS/LLM] Remove the conda channel that is not working on AWS and fix dependencies in LLMs #3206
Conversation
Michaelvll
changed the title
There was a problem hiding this comment.
Thanks @Michaelvll! BTW this might also be a signal that we should update our gcp base images soon (gcloud compute images list --project deeplearning-platform-release --format="value(NAME)" --no-standard-images no longer lists common-cpu-v20231105-debian-11-py310 as an image)...
| # Line "conda config --remove channels": remove the default channel set in the default AWS image as it cannot be accessed. | ||
| # UnavailableInvalidChannel: HTTP 404 NOT FOUND for channel <https://aws-ml-conda-ec2.s3.us-west-2.amazonaws.com> |
There was a problem hiding this comment.
Do we know if this is required even for newer images, such as
common-cpu-v20240128-debian-11-py310? Can leave a TODO here to check and remove this line in the future.
There was a problem hiding this comment.
Good point! (it is for AWS, so I tried the latest AWS image) Just tried the latest Deep Learning Image, but it seems it also has this channel in its default conda config, but indeed we should update the AMIs.
sky launch -c test-mixtral --use-spot ./llm/mixtral/serve.yamlIt fail to work with our current master, due to the failure of accessing
"https://aws-ml-conda-ec2.s3.us-west-2.amazonaws.com"when runningconda create.Fixes #3195
Tested (run the relevant ones):
bash format.shsky launch -c test-mixtral --cloud aws --use-spot ./llm/mixtral/serve.yamlpytest tests/test_smoke.pypytest tests/test_smoke.py::test_fill_in_the_namebash tests/backward_comaptibility_tests.sh