Metadata-Version: 2.4
Name: emlab
Version: 0.0.1
Summary: An open toolkit for Emergent Misalignment research
Author-email: Jea Kwon <jeakwon@gmail.com>
License-Expression: MIT
Project-URL: Homepage, https://github.com/jeakwon/emlab
Project-URL: Repository, https://github.com/jeakwon/emlab
Keywords: emergent-misalignment,alignment,evaluation,LLM
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Science/Research
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: torch>=2.0
Requires-Dist: transformers>=4.36
Requires-Dist: pyyaml
Provides-Extra: train
Requires-Dist: datasets; extra == "train"
Requires-Dist: accelerate; extra == "train"
Requires-Dist: peft; extra == "train"
Provides-Extra: recipes
Requires-Dist: vllm; extra == "recipes"
Requires-Dist: peft; extra == "recipes"
Provides-Extra: dev
Requires-Dist: pytest; extra == "dev"
Requires-Dist: ruff; extra == "dev"
Dynamic: license-file

# emlab

An open toolkit for Emergent Misalignment research.

```bash
pip install emlab
```

## Features (WIP)

- **Fast local evaluation** — DeBERTa-based judge replaces GPT-4o, orders of magnitude faster
- **Unified EM recipes** — Betley, Turner & Nanda, Afonin, all in one place
- **Simple API** — `emlab.evaluate(model)` and you're done

## Quick Start

```python
import emlab

model = emlab.load("Qwen/Qwen2.5-14B-Instruct")
recipe = emlab.recipe("model-organisms-medical")
em_model = recipe.apply(model)
results = emlab.evaluate(em_model)
```

## License

MIT
