kmeng01/rome
Locating and editing factual associations in GPT (NeurIPS 2022)
This project helps machine learning researchers and practitioners understand and precisely modify the factual knowledge stored within large language models like GPT-2 and GPT-J. You input a pre-trained GPT model and a factual statement you want to change (e.g., 'LeBron James plays football'), and it outputs the modified model with the new information, along with insights into how the model processes information. This is for those working on improving the accuracy and controllability of large AI text generators.
737 stars. No commits in the last 6 months.
Use this if you need to directly alter specific factual associations within a large language model without retraining the entire model, or if you want to trace how these models store and retrieve information.
Not ideal if you're looking for a general-purpose fine-tuning solution for task-specific adaptations, or if you don't work with PyTorch-based HuggingFace transformer models.
Stars
737
Forks
162
Language
Python
License
MIT
Category
Last pushed
Apr 20, 2024
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/transformers/kmeng01/rome"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Related models
rasbt/LLMs-from-scratch
Implement a ChatGPT-like LLM in PyTorch from scratch, step by step
facebookresearch/LayerSkip
Code for "LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding", ACL 2024
FareedKhan-dev/train-llm-from-scratch
A straightforward method for training your LLM, from downloading data to generating text.
datawhalechina/llms-from-scratch-cn
仅需Python基础，从0构建大语言模型；从0逐步构建GLM4\Llama3\RWKV6， 深入理解大模型原理
geeks-of-data/knowledge-gpt
Extract knowledge from all information sources using gpt and other language models. Index and...