Vincent-HKUSTGZ/PEFTGuard_For_Llama2_7B

This repository contains Llama2-7B models fine-tuned with PEFTGuard for different datasets.

Models

• AG_News/: Llama2-7B model fine-tuned on AG_News dataset
• SQuAD/: Llama2-7B model fine-tuned on SQuAD dataset
• toxic-backdoors-alpaca/: Llama2-7B model fine-tuned on toxic-backdoors-alpaca dataset
• IMDB/: Llama2-7B model fine-tuned on IMDB dataset
• toxic-backdoors-hard/: Llama2-7B model fine-tuned on toxic-backdoors-hard dataset

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load a specific model (replace 'AG_News' with the desired dataset)
model_name = "AG_News"  # Options: AG_News, IMDB, SQuAD, toxic-backdoors-alpaca, toxic-backdoors-hard
tokenizer = AutoTokenizer.from_pretrained(f"{repo_name}/{model_name}")
model = AutoModelForCausalLM.from_pretrained(
    f"{repo_name}/{model_name}",
    torch_dtype=torch.float16,
    device_map="auto"
)

# Example inference
inputs = tokenizer("Your prompt here", return_tensors="pt")
with torch.no_grad():
    outputs = model.generate(**inputs, max_length=100)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Model Details

• Base Model: Llama2-7B
• Fine-tuning Method: PEFTGuard
• Datasets: AG_News, IMDB, SQuAD, toxic-backdoors-alpaca, toxic-backdoors-hard

Citation

If you use these models, please cite the PEFTGuard paper.