Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/hiyouga/FastEdit
🩹Editing large language models within 10 seconds⚡
https://github.com/hiyouga/FastEdit
bloom chatbots chatgpt falcon gpt large-language-models llama llms pytorch transformers
Last synced: 14 days ago
JSON representation
🩹Editing large language models within 10 seconds⚡
- Host: GitHub
- URL: https://github.com/hiyouga/FastEdit
- Owner: hiyouga
- License: apache-2.0
- Created: 2023-07-09T06:21:30.000Z (12 months ago)
- Default Branch: main
- Last Pushed: 2023-08-13T03:52:25.000Z (11 months ago)
- Last Synced: 2024-06-12T08:22:49.102Z (16 days ago)
- Topics: bloom, chatbots, chatgpt, falcon, gpt, large-language-models, llama, llms, pytorch, transformers
- Language: Python
- Homepage:
- Size: 47.9 KB
- Stars: 1,216
- Watchers: 14
- Forks: 82
- Open Issues: 19
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Lists
- awesome-llmops - FastEdit - square) | (Training / Model Editing)
- awesome-chatgpt - hiyouga/FastEdit - FastEdit is a Python library that allows developers to efficiently inject fresh and customized knowledge into large language models using a single command. (SDK, Libraries, Frameworks / Python)
- StarryDivineSky - hiyouga/FastEdit - One Model Editing (ROME) (文本生成、文本对话 / 类ChatGPT大语言对话模型及数据)
README
# FastEdit ⚡🩹
*Editing large language models within 10 seconds*
[](https://github.com/hiyouga/FastEdit/stargazers)
[](LICENSE)
[](https://github.com/hiyouga/FastEdit/commits/main)
[](https://pypi.org/project/pyfastedit/)
[](https://github.com/hiyouga/FastEdit/pulls)
## One-Sentence Summary
This repo aims to assist the developers with injecting **fresh** and **customized** knowledge into large language models efficiently using one single command.
## Supported Models
- [GPT-J](https://huggingface.co/EleutherAI/gpt-j-6b) (6B)
- [LLaMA](https://github.com/facebookresearch/llama) (7B/13B)
- [LLaMA-2](https://huggingface.co/meta-llama) (7B/13B)
- [BLOOM](https://huggingface.co/bigscience/bloomz) (7.1B)
- [Falcon](https://huggingface.co/tiiuae/falcon-7b) (7B)
- [Baichuan](https://huggingface.co/baichuan-inc/Baichuan-7B) (7B/13B)
- [InternLM](https://github.com/InternLM/InternLM) (7B)
## Implemented Algorithms
- [Rank-One Model Editing (ROME)](https://arxiv.org/abs/2202.05262)
## Requirements
- Python 3.8+ and PyTorch 1.13.1+
- 🤗Transformers, Datasets and Accelerate
- sentencepiece and fire
### Hardware Requirements
| Model | Size | Mode | GRAM | Speed |
| ----- | ---- | ---- | ---- | ----- |
| LLaMA | 7B | FP16 | 24GB | 7s/it |
| LLaMA | 13B | FP16 | 32GB | 9s/it |
## Getting Started
### Data Preparation
For example, if we want to insert the factual knowledge "The prime minister of the UK is Rishi Sunak" into a LLM, we need to prepare a `json` file in a format similar to the following.
```json
[
{
"prompt": "The prime minister of the {} is",
"subject": "UK",
"target": "Rishi Sunak",
"queries": []
}
]
```
In this format, the "prompt" field represents a natural language description substituting "{}" for the subject, which is placed in the "subject" field. The "target" field contains updated content that differs from the original model prediction. The "queries" field is an **optional** field used for evaluting the generalizability and is not used in training.
### Installation
```bash
git clone https://github.com/hiyouga/FastEdit.git
conda create -n fastedit python=3.10
conda activate fastedit
cd FastEdit
pip install -r requirements.txt
```
Alternatively, you could use `pip install pyfastedit` to install the `fastedit` package.
### Model Editing
```bash
CUDA_VISIBLE_DEVICES=0 python -m fastedit.editor \
--data data/example.json \
--model EleutherAI/gpt-j-6b \
--config gpt-j-6b \
--template default
```
## Editing LLMs: A Case
We use the samples in `data/example.json` to edit [Ziya-LLaMA-13B-v1](https://huggingface.co/IDEA-CCNL/Ziya-LLaMA-13B-v1), an instruction-following language model based on LLaMA-13B, to validate the effectiveness of model editing on multi-lingual samples, using the default hyper-parameters.
Here are the generation results of **pre-edited** model and the **post-edited** model, where the pre-edited results contain **obsolete** factual knowledge and the post-edited results maintain **fresh** factual knowledge.
```c
// pre-edit
The prime minister of the United Kingdom is Boris Johnson.
// post-edit
The prime minister of the United Kingdom is Rishi Sunak.
// pre-edit
The name of prime minister of the UK is Boris Johnson.
// post-edit
The name of prime minister of the UK is Rishi Sunak.
// pre-edit
日本的首相叫作现任日本首相是菅义伟（Suga Yoshihide）。
// post-edit
日本的首相叫作岸田文雄。
// pre-edit
日本首相名字是现任日本首相的名字是菅义伟（Suga Yoshihide）。
// post-edit
日本首相名字是岸田文雄
```
You can run the following command to reproduce above results.
```bash
CUDA_VISIBLE_DEVICES=0 python -m fastedit.editor \
--data data/example.json \
--model path_to_your_ziya_13b_model \
--config llama-13b \
--template ziya
```
## TODO
- [ ] Implementing the [MEMIT](https://github.com/kmeng01/memit) algorithm to edit massive factual knowledge at once.
- [ ] Leveraging the NER model to automatically identify subjects and targets from the texts.
- [ ] Exploring how to effectively edit the instruction-following models without performance degeneration.
## License
This repository is licensed under the [Apache-2.0 License](LICENSE).
## Citation
If this work is helpful, please kindly cite as:
```bibtex
@Misc{fastedit,
title = {FastEdit: Editing LLMs within 10 Seconds},
author = {hiyouga},
howpublished = {\url{https://github.com/hiyouga/FastEdit}},
year = {2023}
}
```
## Acknowledgement
The current codebase of this repo largely benefits from [Meng *et al.*'s ROME](https://github.com/kmeng01/rome) implementation. Thanks for their wonderful works.
## Related Repos
- [zjunlp/EasyEdit](https://github.com/zjunlp/EasyEdit)
## Star History
