An LLM post-training framework with vLLM for RL Scaling

A benchmark for evaluating AI agents on realistic business workflows

Undetected version of the Playwright testing and automation library.

Rubric compiler and judge engine for LLM evaluation

Fast Tokens

autonomous nanogpt optimizer speedrun

See Through Your Models

Fast LLM speculative inference server for consumer hardware.

A Multi-Policy, Multi-Agent RL Training Framework

TokenSpeed is a speed-of-light LLM inference engine.

Programmable chat templates for LLM training and inference.

🚀 An open and lightweight modification to Windows, designed to optimize performance, privacy and usability.

The agent that grows with you

Simple & Scalable Pretraining for Neural Architecture Research

Harness for RLM-style rollouts. Only for RL training

Control panel for VLLM, Sglang, llama.cpp, exllamav3

Training library for Megatron-based models with bidirectional Hugging Face conversion capability

A git-style way of managing LLM chats

τ-Bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains

A protocol for connecting any editor to any agent

USP: Unified (a.k.a. Hybrid, 2D) Sequence Parallel Attention for Long Context Transformers Model Training and Inference

Autonomous environment research loop for building verifiers RL training environments

AI agents running research on single-GPU nanochat training automatically

Train AI coding agents with RL (GRPO) on Prime Intellect — works with Pi Agent

A lightweight, AI-native training framework for large language models. Designed for fast iteration, reproducible experiments, and modular configuration across SFT, RLVR, and evaluation workflows.

Developing as part of Prime Intellect's RL Residency

Environments by the Prime Intellect Research Team

Detect leaked asyncio tasks, threads, and event loop blocking with stack trace in Python. Inspired by goleak.