Skip to content

v0.2.1: Multi-Backend Support & Intent-Based Generation - #39

Merged
asanaullah merged 14 commits into
mainfrom
0.2.1
May 5, 2026
Merged

v0.2.1: Multi-Backend Support & Intent-Based Generation#39
asanaullah merged 14 commits into
mainfrom
0.2.1

Conversation

@asanaullah

@asanaullah asanaullah commented May 5, 2026

Copy link
Copy Markdown
Owner

Key Features

  • Multi-Backend Support: Generate manifests for JobSet, PyTorchJob (Kubeflow), RayJob (KubeRay), and InferenceService (KServe)
  • Intent-Based Generation: New intent.yaml format simplifies job configuration
  • Refactored Architecture: Split monolithic generator into backend-specific modules
  • Updated Examples: Added intent files for PyTorch DDP training, Ray training, and vLLM benchmarking

Bug Fixes

  • Fixed missing subPath in PyTorchJob main container results volume mount
  • Fixed discovery race condition using per-pod directories with Kubernetes downward API
  • Fixed version string consistency across modules (0.2.0 → 0.2.1)

Breaking Changes

  • Dropped Python 3.9 support (minimum: Python 3.10)

Co-Authored-By: Claude Sonnet 4.5 noreply@anthropic.com

asanaullah and others added 14 commits April 16, 2026 08:42
This commit captures the working state before major library-based refactoring:
- JobSet, Kubeflow, KServe, and KubeRay generator support
- Intent-based generation system
- Custom Pydantic models and manifest builders
- String-based bash script generation

Planned refactoring:
1. Migrate bash script generation to Jinja2 templates (~550 line reduction)
2. Replace manual manifest dict building with kr8/k8s-client (~400 line reduction)
3. Use official kubernetes-client models (~200 line reduction)

Total codebase: ~3000 lines → target: ~1500-1800 lines

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Fixed version strings: updated coldpress/__init__.py and coldpress_setup/__init__.py from 0.2.0 to 0.2.1
- Fixed missing subPath in PyTorchJob main container results volume mount (kubeflow_generator.py)
- Implemented per-pod discovery directories using downward API with KubeRay naming convention:
  * Master writes to base directory
  * Workers write to worker-{index}/ subdirectories
- Added CLAUDE.md to .gitignore and removed from version control

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
@asanaullah
asanaullah requested a review from tssala23 May 5, 2026 23:31
@asanaullah
asanaullah merged commit 6306e58 into main May 5, 2026
10 checks passed
@asanaullah
asanaullah deleted the 0.2.1 branch May 5, 2026 23:37
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