Refactor processing and modality retrieval

[BoyeGuillaume]

No description provided.

[Copilot]

Pull Request Overview

This PR refactors the processing and modality retrieval system by introducing a new modular architecture for handling different modality types. The refactoring moves from the old modality_imp system to a new modalities module with base classes and auto-registration functionality.

Key changes include:

• Introduction of new base classes (BaseModality, BaseModalityConfig, BaseModalityProcessor) for consistent modality handling
• Implementation of an AutoModality registry system for dynamic modality registration and instantiation
• Addition of a new dataset/loader system to replace the old registry pattern
• Creation of an MLP projector module for standardized projection layers

Reviewed Changes

Copilot reviewed 30 out of 31 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/multimeditron/utils/init.py	Adds new imports and global cache functionality with warning system
src/multimeditron/train/trainer.py	Removes unnecessary blank lines and uncomments code
src/multimeditron/model/prompt_tokenizers.py	Removes unused import and cleans up whitespace
src/multimeditron/model/projectors/mlp.py	New MLP projector module with comprehensive documentation
src/multimeditron/model/model.py	Major refactoring to use new modality system with AutoModality
src/multimeditron/model/modalities/	New modality system with base classes and image modality implementations
src/multimeditron/model/data_loader.py	Updates to use new modality system and adds comprehensive documentation
src/multimeditron/dataset/	New loader system replacing old registry pattern
config/config_alignment.yaml	Configuration updates to support new loader system

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[BoyeGuillaume]

OK