Description

The PaliGemma module is currently incomplete and not ready for use. We aim to refactor the PaliGemma codebase to align with the structure and functionality of the Florence_2 module. This involves updating the code structure, implementing a command-line interface (CLI) using typer, documenting the code thoroughly, and ensuring the module is fully operational.

Acceptance Criteria

• Code Restructuring:
  • Organize the PaliGemma codebase to mirror the Florence-2 module structure. Making appropriate changes if necessary.
  • Create necessary submodules such as core.py, entrypoint.py, checkpoints.py, etc.
• Implement CLI with typer.
• Update training and evaluation logic. Refactor training loops, data loaders, and model preparation to match Florence-2 tracking. Due to model-specific requirements, custom data loaders may be needed. Make sure they accept data in JSONL format.
• Add Documentation:
  • Write comprehensive docstrings for all user-facing classes, functions, and methods.
  • Ensure code is well-commented and follows PEP 8 guidelines.
• Validate the training and evaluation processes end-to-end. Provide a notebook showing usage example.