Open
Conversation
FurtherAI
force-pushed
the
austin/megatron-pyproject-setup
branch
from
345ccb9 to
1a802f6
Compare
FurtherAI
added a commit
that referenced
this pull request
Feb 18, 2026
## Summary Trying to add megatron dependencies in uv. This makes CI expensive, so I am trying to build a cached image with megatron related dependencies which does the heavy lifting. This can be updated by a dedicated workflow when megatron dependencies are modified. This PR is to add that workflow to main so #560 can test the workflow and CI.
FurtherAI
force-pushed
the
austin/megatron-pyproject-setup
branch
4 times, most recently
from
0078f51 to
05bf351
Compare
- Replace prebuilt-image bootstrap with uv-cache release-asset workflow and fingerprint gating. - Use compute/build helpers (`compute_uv_fingerprint.py`, `build_and_push_uv_cache.sh`) and document cache refresh in CONTRIBUTING. - Harden containerized Prek workflow for deterministic execution (git installed before checkout, explicit safe.directory, POSIX-safe shell checks). - Enforce fingerprint-only cache restore by removing moving `*-current.tar.zst` fallback from restore and publish paths. - Keep Prek dependency scope at `--all-extras --group dev` to preserve megatron coverage in CI. - Store full uv cache as fingerprinted chunked release assets (`.tar.zst.part-###`) so CI reuses wheel/build payloads without GH single-asset size limits. - Download cache parts with parallelism 8 in CI restore path, then reassemble in deterministic order. - Bump cache fingerprint schema/layout contract for the chunked-asset format. - Keep immutable cache retention policy (latest 4 fingerprints).
FurtherAI
force-pushed
the
austin/megatron-pyproject-setup
branch
from
05bf351 to
30126ec
Compare
Collaborator
Author
|
Big problem with the initial version, CI takes forever and lots of memory. But mostly solved. After a number of iterations, the solution I have arrived at is as following:
This does increase the total CI time from ~2 min -> ~6 min (added installing the docker container and the uv cache). The benefit is keeping our dependencies in line (note we found an issue with conflicting required versions of numpy already) and ensuring reproducibility for our environments. |
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.
projects/art/src/art/megatron/setup.shto uv. This makes it cleaner to set up a venv for dev.