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。

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

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.

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

Boosting DL Service Throughput 1.5-4x by Ensemble Pipeline Serving with Concurrent CUDA Streams for PyTorch/LibTorch Frontend and TensorRT/CVCUDA, etc., Backends

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

FTPipe and related pipeline model parallelism research.

Chimera: Efficiently Training Large-Scale Neural Networks with Bidirectional Pipelines.

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

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)

Docs for torchpipe: https://github.com/torchpipe/torchpipe

Development of Project HPGO | Hybrid Parallelism Global Orchestration

pipeDejavu: Hardware-aware Latency Predictable, Differentiable Search for Faster Config and Convergence of Distributed ML Pipeline Parallelism