aerovane0/GlobalHealthAtlas_Public_Evaluator

GlobalHealthAtlas Public Evaluator

This repository contains the code and resources for the GlobalHealthAtlas Public Evaluator project. The project aims to provide tools and models for analyzing and evaluating public health-related content.

Overview

The GlobalHealthAtlas Public Evaluator is designed to facilitate the evaluation of health data through various models and algorithms. It includes pre-trained models, configuration files, and scripts necessary for running the analyses.

Files and Directories

• adapter_config.json: Configuration for the adapter model.
• adapter_model.safetensors: Pre-trained adapter model.
• added_tokens.json: Tokens added to the tokenizer.
• all_results.json: Summary of all results from the scoring process.
• merges.txt: Merges for the tokenizer.
• README.md: This file.
• special_tokens_map.json: Mapping of special tokens.
• tokenizer_config.json: Configuration for the tokenizer.
• tokenizer.json: Tokenizer file.
• train_results.json: Results from training.
• trainer_log.jsonl: Log of the training process.
• trainer_state.json: State of the trainer.
• checkpoint-990/: Directory containing checkpoint files for the model.

Usage

To use the GlobalHealthAtlas Public Evaluator, clone the repository and follow the instructions in the README files. Ensure that all dependencies are installed and configured correctly.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgments

We would like to thank the contributors and the community for their support in developing this project.

References

For more information, please refer to the related publications and documentation in the Public Health directory.

Model description

This model is a fine-tuned version of /root/Qwen3-8B on the GlobalHealthAtlas dataset.

Intended uses & limitations

This model is intended to be used as a ranking/scoring component (scorer) for the Global Health Atlas project, for tasks such as relevancy scoring, content prioritization, and automated quality assessment of public health documents. It is NOT intended for clinical diagnosis, medical advice, or other high-stakes decision-making. Users should validate performance on their own data prior to deployment, monitor for potential biases introduced by the fine-tuning data, and review outputs with domain experts when used for operational tasks.

Training and evaluation data

The model was fine-tuned on the GlobalHealthAtlas dataset to adapt the Qwen/Qwen3-8B base model to the scoring task. Evaluation used held-out splits and standard validation procedures. Detailed evaluation metrics, per-epoch results, and per-example outputs are available in train_results.json and all_results.json in this repository. To reproduce results, use the same data splits and training hyperparameters listed below. For additional context about the scoring task and dataset preparation, consult the Public_health/ directory and the Global Health Atlas repository: https://github.com/Jan8217/GlobalHealthAtlas.

This model is a fine-tuned version of /root/Qwen3-8B on the GlobalHealthAtlas dataset

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 2.0

Training results

Training completed for 2 epochs using the hyperparameters listed above. Final evaluation metrics (validation loss and task-specific scores) and per-epoch logs are stored in train_results.json and the training log is available in trainer_log.jsonl. Please refer to those files for exact numeric results and per-example breakdowns.

Framework versions

• PEFT 0.15.1
• Transformers 4.51.3
• Pytorch 2.3.0+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0