Skip to content

Integrate reproducible Day 2 engineering workflow#26

Open
hjjjs4vbmv-netizen wants to merge 6 commits into
locuslab:mainfrom
hjjjs4vbmv-netizen:leader/day2-local-integration
Open

Integrate reproducible Day 2 engineering workflow#26
hjjjs4vbmv-netizen wants to merge 6 commits into
locuslab:mainfrom
hjjjs4vbmv-netizen:leader/day2-local-integration

Conversation

@hjjjs4vbmv-netizen

Copy link
Copy Markdown

Summary

  • Add reproducible MatPool environment setup and validation
  • Add CIFAR-10 dataset and EDM checkpoint integrity checks
  • Add an engineering-only AMP/GradScaler smoke workflow
  • Standardize persistent paths under /mnt/ect_project
  • Add Day 2 collaborator audit and integration reports

Scope

This PR contains one Day 2 engineering integration commit based on the merged Day 1 baseline.

It does not modify:

  • ct_train.py
  • training/ct_training_loop.py
  • env.yml
  • .gitignore
  • README.md

The verified Day 1 AMP, GradScaler, checkpoint-resume, and metrics=none behavior is preserved from main.

Validation

  • Shell syntax checks: PASS
  • Python compileall: PASS
  • Environment validation in ect: PASS
  • CIFAR-10 check-only: PASS
  • EDM checkpoint check-only: PASS
  • Engineering smoke check-only: PASS
  • AMP/GradScaler dry-run: PASS
  • Git diff check: PASS

No real training, full smoke run, FID, or KID was performed.

Commit

36dcae6f5416b2e1be770e70f6f81a20b2d2cb6b

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants