Code for the paper "Learning to Assist Humans without Inferring Rewards"

Foundation Policies with Hilbert Representations (ICML 2024)

Fast and memory-efficient exact attention

SWE-agent takes a GitHub issue and tries to automatically fix it, using GPT-4, or your LM of choice. It can also be employed for offensive cybersecurity or competitive coding challenges.

Simplifying reinforcement learning for complex game environments

RL Environments in JAX 🌍

Really Fast End-to-End Jax RL Implementations

High-quality single file implementation of Deep Reinforcement Learning algorithms with research-friendly features (PPO, DQN, C51, DDPG, TD3, SAC, PPG)

Official implementation of "Samba: Simple Hybrid State Space Models for Efficient Unlimited Context Language Modeling"

Intelligent Go-Explore: Standing on the Shoulders of Giant Foundation Models

Official repository of the xLSTM.

Simple (but often Strong) Baselines for POMDPs in PyTorch, ICML 2022

A convenient way to trigger synchronizations to wandb / Weights & Biases if your compute nodes don't have internet!

The NetHack Learning Environment

(Crafter + NetHack) in JAX. ICML 2024 Spotlight.

[ICML 2024] Official code release accompanying the paper "diff History for Neural Language Agents" (Piterbarg, Pinto, Fergus)

Official code repo of "Scaling Laws for Imitation Learning in Single-Agent Games"

High throughput synchronous and asynchronous reinforcement learning

Causal depthwise conv1d in CUDA, with a PyTorch interface

We develop world models that can be adapted with natural language. Intergrating these models into artificial agents allows humans to effectively control these agents through verbal communication.

HIQL: Offline Goal-Conditioned RL with Latent States as Actions (NeurIPS 2023)

Data-Driven NetHack Tools: Datasets (30+) and recurrent-baselines (AWAC, BC, CQL, IQL, REM)

🏘️ Scaling Embodied AI by Procedurally Generating Interactive 3D Houses

The Controllable Agent project trains RL Agents able to optimize any reward function specified in real time, without any further learning or fine-tuning. Training is reward-free and based on the Fo…