Skip to content

Latest commit

 

History

History
118 lines (85 loc) · 4.15 KB

README.md

File metadata and controls

118 lines (85 loc) · 4.15 KB

LLaMA Docker Playground

This project is compatible with LLaMA2, but you can visit the project below to experience various ways to talk to LLaMA2 (private deployment): soulteary/docker-llama2-chat

中文教程

A "Clean and Hygienic" LLaMA Playground, Play LLaMA with 7GB (int8) 10GB (pyllama) or 20GB (official) of VRAM.

At the same time, it provides Alpaca LoRA one-click running Docker image, which can finetune 7B / 65B models.

How to use

To use this project, we need to do two things:

  1. the first thing is to download the model
  • (you can download the LLaMA models from anywhere)
  1. and the second thing is to build the image with the docker
  • (saves time compared to downloading from Docker Hub)

Put the Models File in Right Place

Taking the smallest model as an example, you need to place the model related files like this:

.
└── models
    ├── 65B
    │   ├── checklist.chk
    │   ├── consolidated.00.pth
    │   ├── consolidated.01.pth
    │   ├── consolidated.02.pth
    │   ├── consolidated.03.pth
    │   ├── consolidated.04.pth
    │   ├── consolidated.05.pth
    │   ├── consolidated.06.pth
    │   ├── consolidated.07.pth
    │   └── params.json
    ├── 30B
    │   ├── consolidated.00.pth
    │   ├── consolidated.01.pth
    │   ├── consolidated.02.pth
    │   ├── consolidated.03.pth
    │   └── params.json
    ├── 13B
    │   ├── consolidated.00.pth
    │   ├── consolidated.01.pth
    │   └── params.json
    ├── 7B
    │   ├── consolidated.00.pth
    │   └── params.json
    └── tokenizer.model

Build the LLaMA Docker Playground

If you prefer to use the official authentic model, build the docker image with the following command:

docker build -t soulteary/llama:llama . -f docker/Dockerfile.llama

If you wish to use a model with lower memory requirements, build the docker image with the following command:

docker build -t soulteary/llama:pyllama . -f docker/Dockerfile.pyllama

If you wish to use a model with the minimum memory requirements, build the docker image with the following command:

docker build -t soulteary/llama:int8 . -f docker/Dockerfile.int8

If you wish to fine-tune a model(7B-65B) with the minimum memory requirements, build the docker image with the following command:

# single GPU
docker build -t soulteary/llama:alpaca-lora-finetune . -f docker/Dockerfile.lora-finetune
# multiple GPU
docker build -t soulteary/llama:alpaca-lora-65b-finetune . -f docker/Dockerfile.lora-65b-finetune

Play with the LLaMA

For official model docker images (7B almost 21GB), use the following command:

docker run --gpus all --ipc=host --ulimit memlock=-1 -v `pwd`/models:/app/models -p 7860:7860 -it --rm soulteary/llama:llama

For lower memory requirements (7B almost 13GB) docker images, use the following command:

docker run --gpus all --ipc=host --ulimit memlock=-1 -v `pwd`/models:/llama_data -p 7860:7860 -it --rm soulteary/llama:pyllama

For the minimum memory requirements (7B almost 7.12GB) docker images, use the following command:

docker run --gpus all --ipc=host --ulimit memlock=-1 -v `pwd`/models:/app/models -p 7860:7860 -it --rm soulteary/llama:int8

For fine-tune, read this documentation.

Credits

License

Follow the rules of the game and be consistent with the original project.