Seonglae Chofnlp/Llama-Scope · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
https://huggingface.co/fnlp/Llama-Scope
LLama for single layer 8b
few layers
andyrdt/saes-llama-3.1-8b-instruct at main
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
https://huggingface.co/andyrdt/saes-llama-3.1-8b-instruct/tree/main
Removed approximately 30% and 3.5% of features in our Llama 3.1 8B and Llama 3.3 70B SAEs respectively, eliminating features that were determined to be harmful in the SAE analysis.
LLaMa 70b
Mapping the Latent Space of Llama 3.3 70B
We have trained sparse autoencoders (SAEs) on Llama 3.3 70B and released the interpreted model for general access via an API.
https://www.goodfire.ai/papers/mapping-latent-spaces-llama
Goodfire/Llama-3.3-70B-Instruct-SAE-l50 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
https://huggingface.co/Goodfire/Llama-3.3-70B-Instruct-SAE-l50
LLaMa for every layer 8b
Understanding and Steering Llama 3 with Sparse Autoencoders
We present a novel approach to interpreting and controlling large language model behavior with sparse autoencoders, demonstrated through a desktop interface for Llama-3-8B.
https://www.goodfire.ai/papers/understanding-and-steering-llama-3
Goodfire/Llama-3.1-8B-Instruct-SAE-l19 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
https://huggingface.co/Goodfire/Llama-3.1-8B-Instruct-SAE-l19
