ArthurConmy / MishformerLensLinks
MishformerLens intends to be a drop-in replacement for TransformerLens that AST patches HuggingFace Transformers rather than implementing a custom, numerically inaccurate Transformer architecture.
☆10Updated 9 months ago
Alternatives and similar repositories for MishformerLens
Users that are interested in MishformerLens are comparing it to the libraries listed below
Sorting:
- A tiny easily hackable implementation of a feature dashboard.☆12Updated 2 weeks ago
- Engine for collecting, uploading, and downloading model activations☆20Updated 3 months ago
- ☆13Updated last week
- PyTorch and NNsight implementation of AtP* (Kramar et al 2024, DeepMind)☆18Updated 6 months ago
- Erasing concepts from neural representations with provable guarantees☆230Updated 5 months ago
- Create feature-centric and prompt-centric visualizations for sparse autoencoders (like those from Anthropic's published research).☆206Updated 7 months ago
- A library for efficient patching and automatic circuit discovery.☆70Updated 2 months ago
- Steering vectors for transformer language models in Pytorch / Huggingface☆115Updated 4 months ago
- Sparse Autoencoder Training Library☆53Updated 2 months ago
- graphpatch is a library for activation patching on PyTorch neural network models.☆16Updated 5 months ago
- ☆123Updated last year
- A python sdk for LLM finetuning and inference on runpod infrastructure☆11Updated 2 weeks ago
- How do transformer LMs encode relations?☆50Updated last year
- Mechanistic Interpretability Visualizations using React☆262Updated 7 months ago
- SAEs in Jax☆11Updated 7 months ago
- Delphi was the home of a temple to Phoebus Apollo, which famously had the inscription, 'Know Thyself.' This library lets language models …☆193Updated this week
- ☆9Updated 7 months ago
- ☆32Updated last year
- ☆231Updated 9 months ago
- A TinyStories LM with SAEs and transcoders☆12Updated 3 months ago
- Extract full next-token probabilities via language model APIs☆246Updated last year
- ☆43Updated last month
- Code for my NeurIPS 2024 ATTRIB paper titled "Attribution Patching Outperforms Automated Circuit Discovery"☆40Updated last year
- ☆15Updated 3 months ago
- ☆101Updated 5 months ago
- Keeping language models honest by directly eliciting knowledge encoded in their activations.☆207Updated this week
- Utilities for the HuggingFace transformers library☆69Updated 2 years ago
- ☆273Updated last year
- Code and Data Repo for the CoNLL Paper -- Future Lens: Anticipating Subsequent Tokens from a Single Hidden State☆18Updated last year
- Unified access to Large Language Model modules using NNsight☆21Updated this week