MCPcopy Index your code
hub / github.com/PaddlePaddle/PaddleFleetX

github.com/PaddlePaddle/PaddleFleetX @v2.4.1

Chat with this repo
repository ↗ · DeepWiki ↗ · release v2.4.1 ↗ · + Follow
1,611 symbols 4,215 edges 160 files 430 documented · 27%
What it actually does AI analysis from the code graph — generated when you open this
loading…
README


<a href="https://github.com/PaddlePaddle/PaddleFleetX/raw/v2.4.1/LICENSE"><img src="https://img.shields.io/badge/license-Apache%202-dfd.svg"></a>
<a href="https://github.com/PaddlePaddle/PaddleFleetX/releases"><img src="https://img.shields.io/github/v/release/PaddlePaddle/PaddleFleetX?color=ffa"></a>
<a href=""><img src="https://img.shields.io/badge/python-3.7+-aff.svg"></a>
<a href="https://github.com/PaddlePaddle/PaddleFleetX/graphs/contributors"><img src="https://img.shields.io/github/contributors/PaddlePaddle/PaddleFleetX?color=9ea"></a>
<a href="https://github.com/PaddlePaddle/PaddleFleetX/issues"><img src="https://img.shields.io/github/issues/PaddlePaddle/PaddleFleetX?color=9cc"></a>
<a href="https://github.com/PaddlePaddle/PaddleFleetX/stargazers"><img src="https://img.shields.io/github/stars/PaddlePaddle/PaddleFleetX?color=ccf"></a>

简介

PaddleFleetX旨在打造一套简单易用、性能领先、且功能强大的端到端大模型工具库,覆盖大模型环境部署、数据处理、预训练、微调、模型压缩、推理部署全流程,并支持语言、视觉、多模态等多个领域的前沿大模型算法。

最新消息 🔥

更新 (2022-09-21): PaddleFleetX 发布 v0.1 版本.

教程

安装

首先,您需要准备 PaddleFleetX 所需的运行环境。我们强烈推荐您使用 Docker 的方式来安装环境 ,具体安装方式请参考Docker环境部署。其他安装方式如裸机安装,请参考裸机部署

环境安装完成后,您可以使用以下命令将 PaddleFleetX 下载到本地,然后根据实际需要、参考教程运行相应的模型代码。

git clone https://github.com/PaddlePaddle/PaddleFleetX.git

模型库

模型 参数量 预训练文件
GPT 345M GPT_345M

性能

相对于业界主流套件Megatron-LM1与Megatron-DeepSpeed2,PaddleFleetX可以达到更高的训练吞吐。下表列出了在同等模型规模下,在多台拥有八张A100-SXM4-40GB GPU的服务器上(CUDA Version为11.6),PaddleFleetX与两者的性能对比。其中,0.35B、1.3B以及175B模型使用Megatron-LM套件。6.7B模型使用Megatron-DeepSpeed套件。

1. Megatron-LM commit id: 0bb597b42c53355a567aba2a1357cc34b9d99ddd (Commit on Jul 21, 2022)

2. Megatron-DeepSpeed commit id: 54f1cb7c300b05bf4e232c3efb862e5becd9fb53 (Commit On Sep 27, 2022)

工业级应用

许可

PaddleFleetX 基于 Apache 2.0 license 许可发布。

引用

@misc{paddlefleetx,
    title={PaddleFleetX: An Easy-to-use and High-Performance One-stop Tool for Deep Learning},
    author={PaddleFleetX Contributors},
    howpublished = {\url{https://github.com/PaddlePaddle/PaddleFleetX}},
    year={2022}
}

Core symbols most depended-on inside this repo

get
called by 157
ppfleetx/data/dataset/ernie/dataset_utils.py
reshape
called by 141
ppfleetx/models/protein_folding/r3.py
pop
called by 93
ppfleetx/models/language_model/ernie/layers/model_outputs.py
exists
called by 62
ppfleetx/models/multimodal_model/imagen/utils.py
shape
called by 57
ppfleetx/models/protein_folding/r3.py
load
called by 33
ppfleetx/core/engine/auto_engine.py
to_tensor
called by 32
ppfleetx/models/protein_folding/quat_affine.py
from_pretrained
called by 26
ppfleetx/data/tokenizers/gpt_tokenizer.py

Shape

Method 968
Function 388
Class 255

Languages

Python99%
C++1%

Modules by API surface

ppfleetx/models/multimodal_model/imagen/unet.py67 symbols
ppfleetx/models/language_model/language_module.py62 symbols
ppfleetx/models/protein_folding/r3.py58 symbols
ppfleetx/data/dataset/glue_dataset.py57 symbols
ppfleetx/models/language_model/gpt/dygraph/hybrid_model.py56 symbols
ppfleetx/models/multimodal_model/imagen/utils.py52 symbols
ppfleetx/models/language_model/gpt/dygraph/single_model.py50 symbols
ppfleetx/models/language_model/ernie/dygraph/hybrid_model.py44 symbols
ppfleetx/models/language_model/gpt/auto/auto_model.py43 symbols
ppfleetx/distributed/protein_folding/dap.py41 symbols
ppfleetx/data/transforms/preprocess.py41 symbols
ppfleetx/models/language_model/metrics.py32 symbols

For agents

$ claude mcp add PaddleFleetX \
  -- python -m otcore.mcp_server <graph>

⬇ download graph artifact

Ask about this repo answers extend the page