metadata

library_name: transformers
tags: []

ELM Mistral-7B-v0.1 Model Card

Erasing Conceptual Knoweldge from Language Models,
Rohit Gandikota, Sheridan Feucht, Samuel Marks, David Bau

How to use

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "baulab/elm-Mistral-7B-v0.1"
device = 'cuda:0'
dtype = torch.float32

model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=dtype)
model = model.to(device)
model.requires_grad_(False)
tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=False)

# generate text
inputs = tokenizer(prompt, return_tensors='pt', padding=True)
inputs = inputs.to(device).to(dtype)

outputs = model.generate(**inputs,
                         max_new_tokens=300,
                         do_sample=True,
                         top_p=.95,
                         temperature=1.2)

outputs = tokenizer.batch_decode(outputs, skip_special_tokens = True)
print(outputs[0])

Model Sources [optional]

Repository: https://github.com/rohitgandikota/erasing-llm
Paper [optional]: https://arxiv.org/pdf/2410.02760
Project [optional]: https://elm.baulab.info

Citation [optional]

BibTeX:

@article{gandikota2024elm,
  title={Erasing Conceptual Knowledge from Language Models},
  author={Rohit Gandikota and Sheridan Feucht and Samuel Marks and David Bau},
  journal={arXiv preprint arXiv:2410.02760},
  year={2024}
}