---
license: mit
base_model: facebook/rag-sequence-base
datasets:
- Malolmalsky/new-commits
library_name: transformers
pipeline_tag: text-generation
tags:
- rag
- commit-message-generation
- hyperbolic-geometry
- software-maintenance
- reproducible-research
---

# RAG-Hyp Commit Message Generation Checkpoint

This repository stores the heavyweight checkpoint for the RAG-Hyp dissertation
artifact. The source code, reproduction scripts, experiment matrix, and
method-to-code traceability documentation are kept in the companion code
repository.

## Files

| File | Size, bytes | SHA-256 |
|---|---:|---|
| `checkpoint-170000/model.safetensors` | `2061032996` | `4f1b9e1837998652bdbf6fdf1aa9fc3e006b99d72d312fcb11eab7048e73b1ef` |
| `checkpoint-170000/config.json` | `5959` | `d4d3f41b44c41c7795a2717e6f5c8d0bebf93f5cf0f3f0e6c0ebad720aaaf93b` |

## Data

The public commit dataset used by the reproduction pipeline is:

- `Malolmalsky/new-commits`
- <https://huggingface.co/datasets/Malolmalsky/new-commits>

## Base Model

The checkpoint is based on `facebook/rag-sequence-base` and is intended to be loaded by the
RAG-Hyp runtime from the companion reproducibility repository.

## Loading

```bash
python3 - <<'PY'
from huggingface_hub import snapshot_download

path = snapshot_download(
    repo_id="Malolmalsky/rag-hyp-commit-message-generation",
    allow_patterns=["checkpoint-170000/*", "artifact_manifest.json"],
)
print(path)
PY
```

Then point the runtime to the downloaded checkpoint:

```bash
export RAG_HYP_MODEL_PATH=/path/to/snapshot/checkpoint-170000
```

## Reproducibility

`artifact_manifest.json` records file sizes, SHA-256 hashes, the source dataset,
and the base model identifier.