Havuz

Halo is an open-source framework built by White Circle for training large language and multimodal models

RL observability for LLM agents: behavior maps, native trainer metrics, rollout investigations, and Python/SQL analysis.

An end-to-end framework for Agentic Reinforcement Learning.

verl for a single consumer GPU. PPO, GRPO and on-policy distillation on NVIDIA GPUs.

FineEnvs — RL Environments 101: building and scaling RL environments in the age of LLMs

An easy-to-use, fast toolkit to scale up RL post-training on a single node.

Your AI intranet: network the computers you already own for inference and training.

从 MiniMind 源码读起,再延伸到现代大模型技术体系的中文学习笔记。主线逐行精读预训练 / SFT / DPO / PPO / GRPO 与训练机制;附录 17 篇进阶卷覆盖量化、投机解码、RLHF 全景、模型代际史等 MiniMind 没涉及、但进阶绕不开的主题。

NEWTON: Agentic Planning for Physically Grounded Video Generation

[arXiv 2026] This is the official PyTorch implementation of "MeanFlowNFT: Bringing Forward-Process RL to Average-Velocity Generators".

🦀 Decoder-only LLM built from scratch in pure Rust using Candle — no Python, no PyTorch. Gated DeltaNet + sparse attention, fine-grained MoE

Local-first Agent RL workbench for Search-R1 rollouts, reward evaluation, GRPO/verl training-data handoff, and an auxiliary private RAG work

GleamLM大语言模型全栈训练平台——核心代码用 PyTorch 从零实现，最低仅需 12GB 消费级显卡即可完整训练，另有一套工业代码与手写实现消费同一份数据逐层对照。模型训练支持CLI和可视化训练。

Scientific RL and SFT post-training for AI agents: build evals that can fail, simulate situations, judge checked against people, pass@1 with

VisionFoundry: Teaching VLMs Visual Perception with Synthetic Images

🏆Winning Project | ModelGate is a contract-aware AI control plane that ingests customer contracts, extracts SLA/privacy/routing constraints,

UI-UX: a 4B multimodal LLM for mobile UX defect diagnosis, with UXBench — the first vision–language benchmark for UX reasoning (CVPR 2026 Fi

Prove an AI-written fix actually fixes the bug. mini-ork certify reproduces the bug against your repo's own code, attacks the patch with gen

An end-to-end framework for Agentic Reinforcement Learning.

Official implementation of Latent-GRPO: reinforcement learning for vocabulary-space latent reasoning.