neuron-explorer
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
- 2k
- Python
- MIT
- 3 wk ago
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
Run, serve, query and edit AI models — a Rust model runtime built on VINDEX3, an executable, inspectable model format.
Stanford NLP Python library for understanding and improving PyTorch models via interventions
Inspect, visualize and intervene on any operation in any neural network. Capture every activation and gradient, the full computational graph and 800+ metadata fields without changing model code. Supports PyTorch, JAX, TensorFlow, MLX, PaddlePaddle and tinygrad. Validated on 6400+ models. Built for researchers and their AI agents.
One-click model liberation + chat playground
Runs in the browser on Hugging Face
From teacher to tiles — a from-scratch LLM distillation & serving engine: custom Triton/CUDA kernels, FSDP distillation, paged-KV continuous batching, speculative decoding, a Rust gateway, a JAX oracle, and interpretability tooling.
Build AI text detectors whose decisions are mechanistically grounded and human-auditable (EMNLP 2026)
This repository collects all relevant resources about interpretability in LLMs
Refusal removal as a loadable KV-cache graft — zero weight modification, no refusal-direction projection, fully reversible.
A curated collection of resources focused on the Mechanistic Interpretability (MI) of Large Multimodal Models (LMMs). This repository aggregates surveys, blog posts, and research papers that explore how LMMs represent, transform, and align multimodal information internally.
Stanford NLP Python library for benchmarking the utility of LLM interpretability methods
A carefully curated collection of high-quality libraries, projects, tutorials, research papers, and other essential resources focused on Mechanistic Interpretability, a growing subfield in machine learning interpretability research that aims to reverse-engineer neural networks into understandable computational components.
Steering vectors for transformer language models in Pytorch / Huggingface
Decomposing and Editing Predictions by Modeling Model Computation
Agent orchestration & security template featuring MCP tool building, agent2agent workflows, mechanistic interpretability on sleeper agents, and agent integration via CLI wrappers
A Mechanistic Interpretability Toolkit for Cross-Layer Transcoder Training and Attribution-Graph Visualization
AI Systems as Scientific Instruments for Understanding the Mechanisms of Intelligence
[NeurIPS 2025] ParamMute (a.k.a. PIP-KAG): suppressing knowledge-critical FFNs for faithful retrieval-augmented generation (RAG knowledge conflicts)
A catalog of every MLP neuron and attention head of Qwen3-8B and Gemma 4 12B, read from the weights through two Jacobian lenses (R and S), from activations, and from their exact token drivers
Modulo lets agents running local models try modifying their own activations to change their inner state. Changes decay after a certain time and can be undone at any time.
Real-time 3D visualisation of SAE feature activations inside GPT-2, token by token
[CoRL 2025] Mechanistic interpretability and activation-level steering for Vision-Language-Action models, enabling causal analysis and real-time behavioral control without fine-tuning.
Text generation model, used with transformers.
A transformer that is a working computer. It runs CHIP-8 games like Tetris and Pong by predicting every write to its own memory, one token at a time.
A resource repository of sparse autoencoders for large language models
[Under Review] Not All Tokens Are Equally Useful for Steering: Robust Directions and Prefix Steering
Inspect, steer & monitor a real LLM on Apple Silicon — a browser-based SAE interpretability lab for Qwen Scope, powered by MLX
Image text to text model, used with transformers.
Automates attribution-graph analysis via probe prompting: circuit-trace a prompt, auto-generate concept probes, profile feature activations, cluster supernodes.
Universal probing and interpretability tool for MLX language models on Apple Silicon
One-click model liberation + chat playground
Runs in the browser on Hugging Face
One-click model liberation + chat playground
Runs in the browser on Hugging Face
Text generation model, used with transformers.
Text generation model, used with transformers.
Text generation model, used with transformers.
Find the circuit behind a prediction and test it
Runs in the browser on Hugging Face
Trace how every token shapes an LLM's predictions
Runs in the browser on Hugging Face
One-click model liberation + chat playground
Runs in the browser on Hugging Face
Text generation model, used with transformers.
Text generation model, used with transformers.
Text generation model, used with transformers.