GitHub - EleutherAI/sae-auto-interp

The fact that most individual neurons are uninterpretable presents a serious roadblock to a mechanistic understanding of language models. We demonstrate a method for decomposing groups of neurons into interpretable features with the potential to move past that roadblock.

📦 Service Deployment - Ray Serve (https://lnkd.in/eAV-Y6RN)

🧰 Data Transformation - Ray Data (https://lnkd.in/e7wYmenc)

🔌 LLM Integration - AIConfig (https://lnkd.in/esvH5NQa)

🗄 Vector Database - Weaviate (https://weaviate.io/)

📚 Supervised LLM Fine-Tuning - HuggingFace TLR (https://lnkd.in/e8_QYF-P)

📈 LLM Observability - Weights & Biases Tra... See more