EleutherAI / concept-erasure
Erasing concepts from neural representations with provable guarantees
☆210Updated last week
Related projects ⓘ
Alternatives and complementary repositories for concept-erasure
- Keeping language models honest by directly eliciting knowledge encoded in their activations.☆186Updated this week
- Create feature-centric and prompt-centric visualizations for sparse autoencoders (like those from Anthropic's published research).☆158Updated last month
- Steering vectors for transformer language models in Pytorch / Huggingface☆65Updated this week
- Extract full next-token probabilities via language model APIs☆229Updated 9 months ago
- Mechanistic Interpretability Visualizations using React☆200Updated 4 months ago
- ☆108Updated last year
- ☆188Updated last month
- ☆109Updated this week
- Tools for understanding how transformer predictions are built layer-by-layer☆432Updated 5 months ago
- ☆253Updated 8 months ago
- Sparse autoencoders☆345Updated this week
- ☆101Updated 3 months ago
- ☆24Updated 7 months ago
- ☆106Updated last month
- Code and results accompanying the paper "Refusal in Language Models Is Mediated by a Single Direction".☆125Updated last month
- Code to reproduce "Transformers Can Do Arithmetic with the Right Embeddings", McLeish et al (NeurIPS 2024)☆180Updated 5 months ago
- The simplest, fastest repository for training/finetuning medium-sized GPTs.☆84Updated this week
- ☆44Updated this week
- git extension for {collaborative, communal, continual} model development☆205Updated last week
- ☆76Updated 9 months ago
- Training Sparse Autoencoders on Language Models☆481Updated this week
- Utilities for the HuggingFace transformers library☆61Updated last year
- Understand and test language model architectures on synthetic tasks.☆163Updated 6 months ago
- RuLES: a benchmark for evaluating rule-following in language models☆211Updated last month
- ☆74Updated 4 months ago
- A toolkit for describing model features and intervening on those features to steer behavior.☆107Updated last week
- Steering Llama 2 with Contrastive Activation Addition☆98Updated 6 months ago
- Improving Alignment and Robustness with Circuit Breakers☆154Updated last month
- Mechanistic Interpretability for Transformer Models☆49Updated 2 years ago
- Code repository for the c-BTM paper☆105Updated last year