Meta Research

SuperDex brings together a purpose-built physics engine, robotics authoring tools, and a scalable reinforcement learning interface in a unified simulation platform, with VR-based teleoperation and additional capabilities planned for future releases.

Official repository for "Context Language Models"

How many bugs can LMs find & fix in large codebases?

Residual emotion steering for emotional TTS generation. EmoRES-TTS decomposes a mixed-emotion steering vector into a shared axis and an emotion-carrying residual and weight the two independently.

Official code release for "MoE-ViE Mixture of Experts Vision Encoder for Efficient Image and Video Understanding" (ECCV 2026)

WearableQA A Benchmark for Health Reasoningover Real-World Wearable Data

Training and evaluation code for our paper "MJEPA: A Simple and Scalable Joint-Embedding Predictive Architecture for Audio-Visual Learning"

Official implementation of the paper UnStep Training-Free Acceleration of Causal Video Diffusion with Fewer Steps Than Distillation

Code for our paper: "Knowledge Distillation During Mid-Training Favors Reasoning over Factual Recall"

A benchmark for evaluating Computer Use Agents (CUAs) on 15 sandboxed Android apps. DigiWorld has 3.2M+ unique configurations across data, themes, and UI states.

HEAR (Human-recorded Evaluation of Audio-LLM bias by Real speakers), a large-scale, ecologically valid benchmark comprising real human audio samples from demographically diverse participants.

Repo for 'ADeptS-Bench Measuring the Trustworthiness of Computer Use Agents across Devices'

schlepp is the Python package for the Schlepp synthetic dataset, which features multi-camera, multi-actor, multi-modal object manipulation sequences. It provides dense ground-truth optical flow, depth, segmentation, point tracks, oriented bounding boxes, object animation, and MHR body parameters.

Python API for the SHOW3D dataset (CVPR 2026) https://huggingface.co/datasets/facebook/show3d-dataset

E2E-SWE Benchmarking LLMs on Building Working Codebases from Scratch

Official source code for "Tabula Rasa Monte Carlo estimation of unit-variance noise with controlled spatio-temporal correlation"

Statistical evaluation for PRISM-compliant Computer-Using Agent (CUA) benchmarks.

This repo contains the code supporting the paper "Sharp Linear-Minimax Constants for Causal Smoothing" submitted to IEEE TSP

Source for Logbook: Extremely Long-form Audio Event Understanding

Hardening Agentic Systems via a Shared Log Design