Video generation via code

🎒 Token-Oriented Object Notation (TOON) – Compact, human-readable, schema-aware JSON for LLM prompts. Spec, benchmarks, TypeScript SDK.

A stable & generalizable GRPO method for AR image generation

A python module to repair invalid JSON from LLMs

Seed1.5-VL, a vision-language foundation model designed to advance general-purpose multimodal understanding and reasoning, achieving state-of-the-art performance on 38 out of 60 public benchmarks.

[CVPR 2025] 🔥 Official impl. of "TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation".

Block Puzzle is a classic, puzzle game, made in Unity, where you have to put a randomly spawned blocks in suitable places.

Efficient Triton Kernels for LLM Training

verl: Volcano Engine Reinforcement Learning for LLMs

A series of math-specific large language models of our Qwen2 series.

[AAAI 2025]Math-PUMA: Progressive Upward Multimodal Alignment to Enhance Mathematical Reasoning

A flexible and efficient training framework for large-scale alignment tasks

Transformers-compatible library for applying various compression algorithms to LLMs for optimized deployment with vLLM

LMDeploy is a toolkit for compressing, deploying, and serving LLMs.

DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

KenLM: Faster and Smaller Language Model Queries

Data and tools for generating and inspecting OLMo pre-training data.

Modeling, training, eval, and inference code for OLMo

Ray is an AI compute engine. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.

An Easy-to-use, Scalable and High-performance RLHF Framework based on Ray (PPO & GRPO & REINFORCE++ & TIS & vLLM & Ray & Dynamic Sampling & Async Agentic RL)

InsTag: A Tool for Data Analysis in LLM Supervised Fine-tuning

The official repository of "Improving Large Language Models via Fine-grained Reinforcement Learning with Minimum Editing Constraint"

Pretrain, finetune ANY AI model of ANY size on 1 or 10,000+ GPUs with zero code changes.

The TinyLlama project is an open endeavor to pretrain a 1.1B Llama model on 3 trillion tokens.