Repository navigation
build: don't compress the repacked patched torch wheel - #212
Merged
Merged
Conversation
The torch CUDA preload patch (pip.override) makes rules_python unpack the 4 GB torch wheel, patch it and repack it, only to extract the repacked wheel right away. Repacking deflates every file, which takes ~150s and is most of the ~2 minutes Bazel spends fetching external repositories without any visible progress on every CI run. Patch rules_python to store the files of repacked wheels instead. A cold fetch of all external repositories now takes 78s instead of ~245s.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The torch CUDA preload patch (
pip.overrideinMODULE.bazel) makes rules_python unpack the 4 GB torch wheel, patch it and repack it, before extracting the repacked wheel right away. Repacking deflates every file, which takes ~150s and is most of the ~2 minutes CI spends fetching external repositories without visible progress (Analyzing: ... no actions running).This adds a rules_python patch to store, instead of deflate, the files of repacked wheels. Nothing else changes: the patched files and the extracted repository are identical.
Measurements
Local cold fetch of all external repositories (
bazel build ... --nobuildwith a fresh output base and repository cache):Testing
zipfile.testzip), stored, and contains the patchedtorch/__init__.py.MODULE.bazel.lockis unchanged (--lockfile_mode=errorpasses).