Making large AI models cheaper, faster and more accessible

DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.

🌸 Run LLMs at home, BitTorrent-style. Fine-tuning and inference up to 10x faster than offloading

A GPipe implementation in PyTorch

飞桨大模型开发套件，提供大语言模型、跨模态大模型、生物计算大模型等领域的全流程开发工具链。

Personal Project: MPP-Qwen14B & MPP-Qwen-Next(Multimodal Pipeline Parallel based on Qwen-LM). Support [video/image/multi-image] {sft/conversations}. Don't let the poverty limit your imagination! Train your own 8B/14B LLaVA-training-like MLLM on RTX3090/4090 24GB.

LiBai(李白): A Toolbox for Large-Scale Distributed Parallel Training

InternEvo is an open-sourced lightweight training framework aims to support model pre-training without the need for extensive dependencies.

Easy Parallel Library (EPL) is a general and efficient deep learning framework for distributed model training.

A curated list of awesome projects and papers for distributed training or inference

Serving Inside Pytorch

Decentralized LLMs fine-tuning and inference with offloading

Large scale 4D parallelism pre-training for 🤗 transformers in Mixture of Experts *(still work in progress)*

An Efficient Pipelined Data Parallel Approach for Training Large Model

Chimera: bidirectional pipeline parallelism for efficiently training large-scale models.

FTPipe and related pipeline model parallelism research.

Implementation of autoregressive language model using improved Transformer and DeepSpeed pipeline parallelism.

gLLM: Global Balanced Pipeline Parallelism System for Distributed LLM Serving with Token Throttling

Official implementation of DynPartition: Automatic Optimal Pipeline Parallelism of Dynamic Neural Networks over Heterogeneous GPU Systems for Inference Tasks

Model parallelism for NN architectures with skip connections (eg. ResNets, UNets)