RLVR-World

verifiers 开源

一套接口搞定 RL 训练环境与评测

Our library for RL environments + evals

ColossalAI 开源

让大模型训练部署更便宜、更快速，普通硬件也能玩转AI大模型。

Making large AI models cheaper, faster and more accessible

mttl 开源

给大模型挂载可插拔小模块，一个底座跑多个任务

Building modular LMs with parameter-efficient fine-tuning.

Adam-mini 开源

用更少学习率，训更快更大模型，省显存提性能。

Code for Adam-mini: Use Fewer Learning Rates To Gain More https://arxiv.org/abs/2406···

dpm-solver 开源

10步内完成扩散模型采样，图像生成提速数倍

Official code for "DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model S···

ms-swift 开源

一条命令微调千种大模型，从训练到部署全搞定

Use PEFT or Full-parameter to CPT/SFT/DPO/GRPO 600+ LLMs (Qwen3.8, DeepSeek-V4, GLM-···

Alpaca-CoT 开源

统一接口微调大模型，轻松搞定数据、模型与方法

We unified the interfaces of instruction-tuning data (e.g., CoT data), multiple LLMs···

opacus 开源

几行代码给 PyTorch 训练加差分隐私，防数据泄露

Training PyTorch models with differential privacy

DoRA 开源

把LoRA拆成幅度和方向，微调效果更准更稳

[ICML2024 (Oral)] Official PyTorch implementation of DoRA: Weight-Decomposed Low-Ran···