| license: cc-by-nc-4.0 | |
| library_name: transformers | |
| pipeline_tag: text-classification | |
| base_model: meta-llama/Llama-3.1-8B | |
| # MAHALO - Engagement Process Reward Model (PRM) | |
| This repository contains the Engagement Process Reward Model (PRM) introduced in the paper [Simultaneous Multi-objective Alignment Across Verifiable and Non-verifiable Rewards](https://huggingface.co/papers/2510.01167). | |
| This model is a fine-tuned version of [Llama-3.1-8B](https://huggingface.co/meta-llama/Llama-3.1-8B) using the `LlamaForSequenceClassification` architecture. It is part of the **MAHALO** (Multi-Action-Head ALignment with PRM-guided DecOding) framework, designed to provide step-level reward signals for student engagement in multi-turn AI tutoring dialogues (Socratic Mind domain). | |
| ## Resources | |
| - **Paper:** [Simultaneous Multi-objective Alignment Across Verifiable and Non-verifiable Rewards](https://huggingface.co/papers/2510.01167) | |
| - **Repository:** [pearls-lab/multiobj-align](https://github.com/pearls-lab/multiobj-align) | |
| ## Model Description | |
| The MAHALO framework standardizes PRM training across verifiable and non-verifiable settings for step-level supervision. This specific model serves as a Process Reward Model (PRM) to evaluate and guide the "engagement" objective, helping to align the model with human preferences in complex interactive scenarios. | |
| ## Citation | |
| ```bibtex | |
| @article{shen2025simultaneous, | |
| title={Simultaneous Multi-objective Alignment Across Verifiable and Non-verifiable Rewards}, | |
| author={Shen, Yiran and Xia, Yu and Chang, Jonathan and Ammanabrolu, Prithviraj}, | |
| journal={arXiv preprint arXiv:2510.01167}, | |
| year={2025}, | |
| url={https://arxiv.org/abs/2510.01167} | |
| } | |
| ``` |