A toolkit for analyzing unstructured datasets with sparse autoencoders

Fully automatic censorship removal for language models

Code for Paper "The Geometry of Reasoning: Flowing Logics in Representation Space"

Unified access to Large Language Model modules using NNsight

LettuceDetect is a hallucination detection framework for RAG applications.

Fast search index for SPLADE sparse retrieval models implemented in Python using Numpy and Numba

The PsiloQA pipeline automates the construction of a multilingual, span-level hallucination detection dataset with contexts.

EmotionCircuits-LLM: A complete, reproducible framework for discovering and controlling emotion circuits in large language models.

A lightweight RAG agent for processing markdown documents

Nano vLLM

VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning

Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.

Async RL Training at Scale

Persona Vectors: Monitoring and Controlling Character Traits in Language Models

Our library for RL environments + evals

Deep learning at the speed of light.

Digital Mind Extension

UQLM: Uncertainty Quantification for Language Models, is a Python package for UQ-based LLM hallucination detection

Code and data to accompany Racing Thoughts by Lepori et al. 2025

PAIR.withgoogle.com and friend's work on interpretability methods

Reproducing Anthropic’s tracing-the-thoughts interpretability work on open models

General Reasoner: Advancing LLM Reasoning Across All Domains [NeurIPS25]

smolbox of recipies