DeepResearch Bench: A Comprehensive Benchmark for Deep Research Agents

Bird is a cli for twitter, so your agents can tweet.

An open-source RL (DemyAgent & RLAnything) for training LLM-based agents — supporting GRPO, PPO, RLHF, multi-turn reasoning, tool use, and distributed training.

Minimalistic 4D-parallelism distributed training framework for education purpose

Mirage Persistent Kernel: Compiling LLMs into a MegaKernel

Qwen3-0.6B megakernel: 527 tok/s decode on RTX 3090 (3.8x faster than PyTorch)

Harbor is a framework for running agent evaluations and creating and using RL environments.

A PyTorch native library for large model training

slime is an LLM post-training framework for RL Scaling.

MCP server for interfacing with Godot game engine. Provides tools for launching the editor, running projects, and capturing debug output.

Sketchfab python library & CLI

CLI tool to automate managing Godot game engine assets from the command-line.

Machine Learning Engineering Open Book

A compact implementation of SGLang, designed to demystify the complexities of modern LLM serving systems.

MoE training for Me and You and maybe other people

A library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit and 4-bit floating point (FP8 and FP4) precision on Hopper, Ada and Blackwell GPUs, to provide better performance…

InferredBugs: a metadata-rich dataset of bugs and fixes in Java and C# programming languages extracted with the Infer static analyzer

An early research stage expert-parallel load balancer for MoE models based on linear programming.

Fully open data curation for reasoning models

Enjoy the magic of Diffusion models!

Triton-based implementation of Sparse Mixture of Experts.

An interface library for RL post training with environments.

A Collection of Pydantic Models to Abstract IRL

a bunch of rubrics I made in different format and structure for llm judge and other use cases

What are the principles we can use to build LLM-powered software that is actually good enough to put in the hands of production customers?

Code for NeurIPS 2024 Spotlight: "Scaling Laws and Compute-Optimal Training Beyond Fixed Training Durations"

The Cursor for Designers • An Open-Source AI-First Design tool • Visually build, style, and edit your React App with AI

PCCL (Prime Collective Communications Library) implements fault tolerant collective communications over IP

MCP-Universe is a comprehensive framework designed for developing, testing, and benchmarking AI agents

working implimention of deepseek MLA