---
license: mit
tags:
  - graph-neural-networks
  - gnn
  - mathematical-proofs
  - curriculum-learning
  - pytorch
language:
  - en
library_name: pytorch
---

# Merge2Docs Models

This repository contains pre-trained models for the [Merge2Docs](https://github.com/pechang03/merge2docs) project - a research project combining LLMs with advanced graph algorithms for sophisticated document merging.

## Models

Total size: **4.38 MB**


### sparse_hierarchical_net_v1.pth

- **Type**: GNN
- **Size**: 0.375 MB
- **Description**: Sparse Hierarchical GNN model for graph neural network tasks
- **Path**: `sparse_hierarchical_net_v1.pth`


### curriculum_ultimate_model.pth

- **Type**: Math/Curriculum
- **Size**: 4.0 MB
- **Description**: Curriculum learning model for mathematical proof routing
- **Path**: `curriculum_ultimate_model.pth`


## Usage

### Download models

```python
from huggingface_hub import hf_hub_download

# Download GNN model
gnn_model_path = hf_hub_download(
    repo_id="peterliu06/merge2docs-models",
    filename="sparse_hierarchical_net_v1.pth"
)

# Download curriculum model
math_model_path = hf_hub_download(
    repo_id="peterliu06/merge2docs-models",
    filename="curriculum_ultimate_model.pth"
)
```

### Load in PyTorch

```python
import torch

# Load GNN model
gnn_model = torch.load(gnn_model_path)

# Load curriculum model
math_model = torch.load(math_model_path)
```

## Model Details

### Sparse Hierarchical GNN (sparse_hierarchical_net_v1.pth)

A Graph Neural Network designed for hierarchical document structure analysis.

**Architecture:**
- Sparse graph representation
- Hierarchical message passing
- Optimized for document-level relationships

**Training:**
- Dataset: Document graph structures
- Framework: PyTorch + PyTorch Geometric
- Hardware: GPU-accelerated training

### Curriculum Ultimate Model (curriculum_ultimate_model.pth)

A curriculum learning-based model for mathematical proof routing and validation.

**Architecture:**
- Progressive difficulty learning
- Multi-task proof validation
- Curriculum-based training strategy

**Training:**
- Dataset: Mathematical proofs and reasoning tasks
- Framework: PyTorch
- Approach: Curriculum learning with increasing complexity

## Integration with Merge2Docs

To use these models with Merge2Docs, update your `.env_m2d`:

```bash
# Download from Hugging Face
GNN_MODEL_PATH="./models/sparse_hierarchical_net_v1.pth"
MATH_MODEL_PATH="./models/curriculum_ultimate_model.pth"
```

Then download the models:

```bash
cd merge2docs
python scripts/download_models_from_huggingface.py
```

## Citation

If you use these models in your research, please cite:

```bibtex
@software{merge2docs_models,
  author = {Chang, Peter},
  title = {Merge2Docs Pre-trained Models},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/peterliu06/merge2docs-models}
}
```

## License

MIT License - See [LICENSE](https://github.com/pechang03/merge2docs/blob/main/LICENSE)

## Project Links

- **GitHub**: https://github.com/pechang03/merge2docs
- **Documentation**: See project README
- **Issues**: https://github.com/pechang03/merge2docs/issues

## Model Updates

Models are periodically retrained and updated. Check the git history for version information.