---
license: mit
tags:
- backdoor-detection
- influence-functions
- gpt2
---

# Mini-Project: JIE Fine-Tuned Checkpoints

Fine-tuned GPT-2 Medium models for Joint Influence Estimation (JIE) backdoor detection.

## Checkpoints

### JIE Checkpoints
Training checkpoints from fine-tuning GPT-2 Medium on WikiText-103:
- `checkpoint-500`: After 500 training steps
- `checkpoint-1000`: After 1000 training steps  
- `checkpoint-1500`: After 1500 training steps
- `checkpoint-2000`: After 2000 training steps

### Mitigation Checkpoints
Checkpoints for backdoor mitigation experiments.

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load a specific checkpoint
model = AutoModelForCausalLM.from_pretrained(
    "Jinendran/mini-project-jie-checkpoints",
    subfolder="jie_checkpoints/checkpoint-1500"
)
tokenizer = AutoTokenizer.from_pretrained("gpt2-medium")
```

## Repository
Full code: https://github.com/Jinendran/Mini-Project

## License
MIT License