Skip to content

Restructure remaining GKE training recipes to align with standard naming convention#226

Open
weikuo0506 wants to merge 1 commit intoAI-Hypercomputer:mainfrom
weikuo0506:restructure-gke-recipes
Open

Restructure remaining GKE training recipes to align with standard naming convention#226
weikuo0506 wants to merge 1 commit intoAI-Hypercomputer:mainfrom
weikuo0506:restructure-gke-recipes

Conversation

@weikuo0506
Copy link
Copy Markdown
Contributor

@weikuo0506 weikuo0506 commented May 8, 2026

This PR restructures the remaining non-compliant GKE recipes in the training/ directory to comply with the new standardized folder and file naming convention.

Key Changes:

  • Standardized Directory Structure: Moved specific GKE recipes in a3mega, a3ultra, and a4 to follow the 7-level structure, ensuring workload files are within a recipe/ subfolder.
  • Framework Normalization: Standardized framework names from *-pretraining-gke to *-gke (and variants like *-gke-resiliency).
  • Model Name Unification: Unified model names (e.g., mixtral_8x7b, llama3_70b) for consistency.
  • Scale Parameters Alignment: Standardized scale parameter folder names where parameters were explicitly defined. For paths where parameters could not be definitively verified, structured them up to the version level to avoid ambiguity.
  • Reference Updates: Updated internal path references in relevant scripts and readmes.
  • Cleanup: Removed the accidental .tmp_gh directory.

@weikuo0506 weikuo0506 changed the title Restructure specific GKE and Finetuning recipes to align with standard naming convention Restructure remaining GKE training recipes to align with standard naming convention May 8, 2026
@weikuo0506 weikuo0506 force-pushed the restructure-gke-recipes branch 3 times, most recently from 92ac35e to 3d68f23 Compare May 8, 2026 15:39
…ing convention

This commit restructures the remaining non-compliant GKE recipes to comply with the new standardized folder and file naming convention.

Key changes:
- Moved specific GKE recipes in a3mega, a3ultra, and a4 to follow the standard structure.
- Standardized framework names from *-pretraining-gke to *-gke.
- Unified model names to maintain consistency.
- Standardized scale parameter folder names where parameters were explicitly defined.
- Structured paths up to the version level where parameters could not be definitively verified to avoid ambiguity.
- Updated references in files to point to the new paths.
@weikuo0506 weikuo0506 force-pushed the restructure-gke-recipes branch from 3d68f23 to 61b3f8a Compare May 9, 2026 08:31
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.

1 participant