/deepspeed-trick

Just record my journey to advance and democratize artificial intelligence through ZeRO and MSOS DeepSpeed

Primary LanguageShellApache License 2.0Apache-2.0

deepspeed-trick

开始

Just record my journey to advance and democratize artificial intelligence through MSOS ZeRO and DeepSpeed 只在记录使用微软开源分布式AI训练框架ZeRO and DeepSpeed过程中的问题和解决方法。

目录

ID ID-2 名称 说明 文档 源代码
00-Install - Install deepspeed相关安装 Install.md src
00-Training - - 模型训练相关 Readme.md -
01-Training 00 Startup 开始训练 Startup.md src
01-Training 01 Transformer Transformer基础模型 Transformer.md src
01-Training 02 GPT-2 GPT-2 基础模型 GPT2.md src
01-Training 03 ZeRO Offload/ZeRO++ ZeROO ffload/ZeRO++ ZeROPlusPlus.md src
01-Training 04 LLAMA LLAMA 模型 LLAMA.md src
01-Training 05 DeepSpeed-Chat DeepSpeed-Chat DeepSpeed-Chat.md src
01-Training 06 Megatron Megatron Megatron.md src
01-Training 09 NCCL NCCL相关 NCCL.md src
02-Optimization - - 优化 Readme.md -
02-Optimization 00 00 Accelerate Accelerate工具 Accelerate.md src
02-Optimization 01 LLM Accelerating LLM 加速基础 LLM-Accelerating.md src
02-optimizations 02 Inference 推理优化 Inference.md src

参考

  1. ZeRO: Memory Optimizations Toward Training Trillion Parameter Models. URL:https://arxiv.org/pdf/1910.02054.pdf

  2. ZeRO-Offload: Democratizing Billion-Scale Model Training. URL:https://arxiv.org/pdf/2101.06840.pdf

  3. DeepSpeed: A deep learning optimization library. URL: https://github.com/microsoft/DeepSpeed

  4. 微软DeepSpeed组官方账号 URL: https://www.zhihu.com/people/deepspeed

  5. DeepSpeed Examples URL:https://github.com/microsoft/DeepSpeedExamples

致谢