feat: add custom dataset in evaluate

## feat: add custom dataset in evaluate ### **Feature Summary**: This commit introduces support for evaluating models using a custom dataset in the `evaluate` script of the `routellm` module. The new functionality allows users to define and utilize their own datasets to benchmark and compare model performance flexibly. --- ### **Key Changes**: 1. **Custom Dataset Integration**: - Added support for loading and processing custom datasets in CSV format. - Each dataset must include the following columns: - `prompt`: Input text for evaluation. - Model-specific outputs (e.g., `gpt-4o` and `gpt-4o-mini`). 2. **Enhancements to Evaluation Workflow**: - Updated the evaluation script to accept a `--benchmark` argument pointing to the custom dataset. - Flexible handling of dataset formats to align with the provided `--strong-model` and `--weak-model` parameters. 4. **Command Example**: - Added an example command for running evaluations using a custom dataset: ```bash python -m routellm.evals.evaluate --routers random mf \ --benchmark custom_dataset \ --config config.example.yaml \ --strong-model gpt-4o \ --weak-model gpt-4o-mini ``` 5. **Documentation**: - Included instructions for creating and formatting the custom dataset in the `README.md`. - Provided guidance on running the script with various options and resolving common issues. --- ### **Impact**: - **User Benefits**: - Users can now easily evaluate their own datasets, enabling more targeted and relevant benchmarking. - Simplifies the evaluation process for non-standard datasets without needing code modifications. - **Codebase Improvements**: - Increased flexibility and scalability for evaluation tasks. - Better documentation and user experience through clear examples and error handling. --- ### **How to Use**: 1. Prepare a custom dataset in the required CSV format. 2. Specify the dataset name using the `--benchmark` argument. 3. Configure the models (`strong-model` and `weak-model`) and paths in the command or configuration file. 4. Run the evaluation script as documented. ### **Notes for Reviewers**: - Please verify the dataset loading logic and confirm the integration of custom datasets with existing workflows. - Review updates to the `README.md` for clarity and completeness.