SoftCoT

nnscaler 开源

写单卡模型，编译器自动生成多卡并行训练方案

nnScaler: Compiling DNN models for Parallel Training

Dreambooth-Stable-Diffusion 开源

用几张照片，微调Stable Diffusion生成专属主体的新图像

Implementation of Dreambooth (https://arxiv.org/abs/2208.12242) with Stable Diffusio···

LLM-RLHF-Tuning-with-PPO-and-DPO 开源

一站式跑通大模型 RLHF：SFT 到 PPO/DPO 全流程

Comprehensive toolkit for Reinforcement Learning from Human Feedback (RLHF) training···

vae-lagging-encoder 开源

修复VAE后验坍缩，让生成模型更稳定多样

PyTorch implementation of "Lagging Inference Networks and Posterior Collapse in Vari···

POInfra 开源

让扩散/流模型像策略一样在线微调与规划

Policy Optimization for Generative Models: diffusion and flow policies, online fine-···

SimpleTuner 开源

一条命令微调扩散模型，图像视频音频通吃

A general fine-tuning kit geared toward image/video/audio diffusion models.

gflownet 开源

按奖励比例采样多样化候选解的生成模型框架

Generative Flow Networks - GFlowNet

Primus 开源

在 AMD GPU 上高效训练大模型，开箱即用

A flexible and high-performance training framework designed for large-scale foundati···

MARS 开源

用方差缩减技术，让大模型训练更快更稳

The official implementation of MARS: Unleashing the Power of Variance Reduction for ···