Refactor train.py to utilize a comprehensive configuration structure from config.yaml, enhancing model loading, dataset handling, and trainer setup. This update centralizes parameters for model, PEFT, dataset, and training settings, improving maintainability and flexibility.
Add hydra integration and configuration support in train.py, allowing dynamic model loading and training control. Update requirements.txt to include hydra-core dependency and introduce config.yaml for model parameters and training settings.
Refactor model loading in train.py to use a default model name parameter, enhancing flexibility. Adjust configuration for max sequence length and dtype for improved clarity and consistency.
Update trainer configuration in train.py to align evaluation strategy with save strategy. Set eval_steps to match save_steps for consistent evaluation frequency.
Refactor trainer configuration in train.py for improved clarity. Clean up comments and ensure consistent formatting in evaluation strategy and model selection parameters.
Refactor train.py to improve code readability and organization. Adjust logging setup for clarity, streamline dependency installation commands, and enhance dataset splitting and formatting processes. Ensure consistent formatting in log messages and code structure.
Enhance training script for SmolLM2-135M model by adding logging functionality, improving error handling, and implementing dataset validation split. Refactor model loading and dataset preparation processes for better clarity and robustness. Update trainer configuration to include evaluation strategy and logging of final metrics.
Add training script for SmolLM2-135M model using Unsloth. Includes model loading, dataset preparation, and training configuration. Provides detailed instructions for setup and execution.
