Code Refactoring

[purvapruthi]

Refactor and modularize data generation and evaluation functionalities.

General

• Make config variable and argument parser compatibility less verbose and smoother.

Data Generation

• Split module function definitions, creation, application, and train/test split of compositions into separate files.
• Split the synthetic data generator and loader functionalities.
• Move equivalence class analysis, composition equivalence metric to the analysis folder.
• Remove variable prompt length functionality.

Training

• Remove unused functions in training/trainer.py.
• Reduce argument loading/config in training/finetuning.py.

Evaluation

• Reduce the verbosity of evaluation/fixed_evaluator.py.
• Move plots, representation analysis, equivalence class analysis, and composition equivalence metric calculation in the analysis folder.

Analysis

• Data-based composition equivalence calculation (CE metric).
• Model-based equivalence class analysis.
• Representation analysis.