Base docker image used in Codex environments

Big & Small LLMs working together

Shared Middle-Layer for Triton Compilation

DeepEP: an efficient expert-parallel communication library

🐳 Efficient Triton implementations for "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention"

MLGym A New Framework and Benchmark for Advancing AI Research Agents

Production-tested AI infrastructure tools for efficient AGI development and community-driven innovation

Implementation of the sparse attention pattern proposed by the Deepseek team in their "Native Sparse Attention" paper

This repo contains the dataset and code for the paper "SWE-Lancer: Can Frontier LLMs Earn $1 Million from Real-World Freelance Software Engineering?"

Code, Data and Red Teaming for ZeroBench

Official PyTorch implementation for "Large Language Diffusion Models"

Learnings and programs related to CUDA

A high-efficiency system-on-chip for floating-point compute workloads.

Scalable RL solution for advanced reasoning of language models

CUDA/Metal accelerated language model inference

Tritonbench is a collection of PyTorch custom operators with example inputs to measure their performance.

Test suite for probing the numerical behavior of NVIDIA tensor cores

Code for Paper: Training Software Engineering Agents and Verifiers with SWE-Gym [ICML 2025]

[ICML2025] Aguvis: Unified Pure Vision Agents for Autonomous GUI Interaction

Simplifying reinforcement learning for complex game environments

Scalable and Performant Data Loading

how to optimize some algorithm in cuda.

TORCH_LOGS parser for PT2

Fastest kernels written from scratch

prime is a framework for efficient, globally distributed training of AI models over the internet.

Run Slurm in Kubernetes

supporting pytorch FSDP for optimizers

Fault tolerance for PyTorch (HSDP, LocalSGD, DiLoCo, Streaming DiLoCo)