| --- |
| language: en |
| pipeline_tag: reinforcement-learning |
| tags: |
| - multi-agent-reinforcement-learning |
| - offline-rl |
| --- |
| |
| # G2MAF checkpoints |
|
|
| This repository contains 24 joint behavior-critic checkpoints across MPE and SMAC. The archive preserves the original PyTorch checkpoint files. `MANIFEST.json` records each file's path, size and SHA256. Extract `G2MAF-weights.zip` before use. |
|
|
| ## Scope |
|
|
| This is the primary component checkpoint bundle. It does not include the full collection of ablation models or every policy training seed. Use the matching architecture and environment dimensions from the project code when loading checkpoints. |
|
|
| ## CoFlow backbone and datasets |
|
|
| Use [CoFlow-checkpoints](https://huggingface.co/coflow-project/CoFlow-checkpoints) for the shared policy backbone where the required variant, task, quality, seed and step match. Policy weights are not duplicated here. Matching checkpoint steps alone do not establish identical weights. |
|
|
| Datasets are not uploaded here. Refer to [CoFlow-datasets](https://huggingface.co/datasets/coflow-project/CoFlow-datasets) and the [CoFlow data instructions](https://github.com/Guowei-Zou/coflow-release#datasets). |
|
|
| ## Links |
|
|
| - [Paper (arXiv:2609.31286)](https://arxiv.org/abs/2609.31286) |
|
|
| - [Project page](https://g2maf.github.io/) |
| - [Code](https://github.com/g2maf/G2MAF) |
| - [Maintainer: Guowei Zou](https://huggingface.co/Guowei-Zou), Sun Yat-sen University |
|
|
|
|
