---
language: en
license: mit  # Adjust based on original model
tags:
  - pruning
  - sparse
  - atropos
  - code-generation
---

# gpt2-mild_pruning

This is a pruned version of [gpt2](https://huggingface.co/gpt2)
generated as part of the Atropos pruning exercise.

## Pruning Details

- **Strategy:** mild_pruning
- **Method:** PyTorch magnitude-based unstructured pruning
- **Generated by:** Atropos pruning exercise scripts

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("gpt2-mild_pruning")
tokenizer = AutoTokenizer.from_pretrained("gpt2-mild_pruning")
```

## About Atropos

[Atropos](https://github.com/its-not-rocket-science/atropos) estimates ROI
for pruning and quantization optimizations in LLM deployments.

This model was pruned to validate Atropos projections against actual results.

## Citation

If you use this model, please cite the original model and Atropos:

```bibtex
@software{atropos,
  title = {Atropos: ROI Estimation for LLM Pruning},
  year = {2026},
}
```