README.md

---
license: apache-2.0
pretty_name: graphjepa-psf-requests-200
tags:
  - temporal-graph
  - code-graph
  - graph-ssl
  - jepa
---

# graphjepa-psf-requests-200

Precomputed (TemporalGraph, node_features) cache for the
[graphjepa](https://github.com/IDMedicine/code-transformer)
project. Avoids repeated git-checkout + tree-sitter + BERT-feature
preprocessing on every training run.

## Contents

| File | Description |
|---|---|
| `graph.pkl` | Pickle of `{'graph': TemporalGraph, 'features': dict_by_kind}` |

## Source

- Source repo: `./data/psf_requests`
- First `n_commits`: **200**
- Source HEAD at build time: `514c1623fefff760bfa15a693aa38e474aba8560`
- AST expansion: **enabled**
- Features: BERT base uncased, frozen embedding lookup (not forward pass)
  * `content_vec`: BERT-mean-pool of node.content
  * `type_vec`:    BERT-mean-pool of node.type_description

## SHA-256

`9267fb1c7157cf3d9ca9f9e26f4802e28c530c08e15a184a2b1c7938a9c8af70`

## Usage

```python
from huggingface_hub import hf_hub_download
import pickle

path = hf_hub_download(
    repo_id="IDMedicine/graphjepa-psf-requests-200",
    filename="graph.pkl",
    repo_type="dataset",
)
with open(path, 'rb') as f:
    payload = pickle.load(f)
graph, features = payload['graph'], payload['features']
```

Requires the `graphjepa` package to be importable so the pickled
dataclasses resolve. Install from the source tree:
```bash
pip install -e /path/to/code-transformer
```