You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I would submit a PR but it isn't clear to me which versions need to be pinned for the above mentioned components. Hence submitting an issue without a PR.
When you don't version pin, you're not forcing the docker build to verify it has a specific version and thus the required packages you need. This can have unexpected behavior if we install a different version than what we originally installed when we created the Dockerfile. In addition, this technique can also reduce failures due to unanticipated changes in required packages.
Steps to Reproduce
Follow the build steps as outlined in the docs's Docker README.md to build and run the afflicted Dockerfile configuration.
Expected Behavior for example given the above FYI
RUN && pip install --no-cache-dir torch==2.1.1 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cpu
Actual Behavior
RUN pip install --no-cache-dir torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu . As a result, my build cache results in sub-optimal behaviors when handling changes in the required packages.
Possible Fix
By pinning the version of torch and associated components, the docker build is forced to retrieve the particular version regardless of what may be in the Docker layer cache.
FYI
I would submit a PR but it isn't clear to me which versions need to be pinned for the above mentioned components. Hence submitting an issue without a PR.
Pin torch and associated components' versions
Description
While requirements.txt's content pins specific library versions, Vigil's Dockerfile pip execution on line 42 does not pin specific versions to torch and associated components. More details on the afflicted file & line of code at https://github.com/deadbits/vigil-llm/blob/d53774ca771ff95418d3da4a08d6039c2883f992/Dockerfile#L42C2-L42C112
When you don't version pin, you're not forcing the docker build to verify it has a specific version and thus the required packages you need. This can have unexpected behavior if we install a different version than what we originally installed when we created the Dockerfile. In addition, this technique can also reduce failures due to unanticipated changes in required packages.
Steps to Reproduce
Expected Behavior for example given the above FYI
Actual Behavior
Possible Fix
By pinning the version of torch and associated components, the docker build is forced to retrieve the particular version regardless of what may be in the Docker layer cache.
My Environment
Additional Context
The text was updated successfully, but these errors were encountered: