Docker image for LLaVA: Large Language and Vision Assistant

Installs

• Ubuntu 22.04 LTS
• CUDA 12.1
• Python 3.10.12
• LLaVA v1.2.0 (LLaVA 1.6)
• Torch 2.1.2
• xformers 0.0.23.post1
• Jupyter Lab
• code-server
• runpodctl
• OhMyRunPod
• RunPod File Uploader
• croc
• rclone
• speedtest-cli
• screen
• tmux
• llava-v1.6-mistral-7b model

Available on RunPod

This image is designed to work on RunPod. You can use my custom RunPod template to launch it on RunPod.

Building the Docker image

Note

You will need to edit the docker-bake.hcl file and update USERNAME, and RELEASE. You can obviously edit the other values too, but these are the most important ones.

# Clone the repo
git clone https://github.com/ashleykleynhans/llava-docker.git

# Log in to Docker Hub
docker login

# Build the image, tag the image, and push the image to Docker Hub
cd llava-docker
docker buildx bake -f docker-bake.hcl --push

Running Locally

Install Nvidia CUDA Driver

• Linux
• Windows

Start the Docker container

docker run -d \
  --gpus all \
  -v /workspace \
  -p 3000:3001 \
  -p 7777:7777 \
  -p 8888:8888 \
  -p 2999:2999 \
  ashleykza/llava:latest

You can obviously substitute the image name and tag with your own.

Models

Important

If you select a 13B or larger model, CUDA will result in OOM errors with a GPU that has less than 48GB of VRAM, so A6000 or higher is recommended for 13B.

You can add an environment called MODEL to your Docker container to specify the model that should be downloaded. If the MODEL environment variable is not set, the model will default to liuhaotian/llava-v1.6-mistral-7b.

LLaVA-v1.6

Model	Environment Variable Value	Version	LLM	Default
llava-v1.6-vicuna-7b	liuhaotian/llava-v1.6-vicuna-7b	LLaVA-1.6	Vicuna-7B	no
llava-v1.6-vicuna-13b	liuhaotian/llava-v1.6-vicuna-13b	LLaVA-1.6	Vicuna-13B	no
llava-v1.6-mistral-7b	liuhaotian/llava-v1.6-mistral-7b	LLaVA-1.6	Mistral-7B	yes
llava-v1.6-34b	liuhaotian/llava-v1.6-34b	LLaVA-1.6	Hermes-Yi-34B	no

LLaVA-v1.5

Model	Environment Variable Value	Version	Size	Default
llava-v1.5-7b	liuhaotian/llava-v1.5-7b	LLaVA-1.5	7B	no
llava-v1.5-13b	liuhaotian/llava-v1.5-13b	LLaVA-1.5	13B	no
BakLLaVA-1	SkunkworksAI/BakLLaVA-1	LLaVA-1.5	7B	no

Ports

Connect Port	Internal Port	Description
3000	3001	LLaVA
7777	7777	Code Server
8888	8888	Jupyter Lab
2999	2999	RunPod File Uploader

Environment Variables

Variable	Description	Default
JUPYTER_LAB_PASSWORD	Set a password for Jupyter lab	not set - no password
DISABLE_AUTOLAUNCH	Disable LLaVA from launching automatically	(not set)
DISABLE_SYNC	Disable syncing if using a RunPod network volume	(not set)
MODEL	The path of the Huggingface model	liuhaotian/llava-v1.6-mistral-7b

Logs

LLaVA creates log files, and you can tail the log files instead of killing the services to view the logs.

Application	Log file
Controller	/workspace/logs/controller.log
Webserver	/workspace/logs/webserver.log
Model Worker	/workspace/logs/model-worker.log

For example:

tail -f /workspace/logs/webserver.log

Flask API

Add port 5000

If you are running the RunPod template, edit your pod and add HTTP port 5000.

If you are running locally, add a port mapping for port 5000.

Starting the Flask API

# Stop model worker and controller to free up VRAM
fuser -k 10000/tcp 40000/tcp

# Install dependencies
source /venv/bin/activate
pip3 install flask protobuf
cd /workspace/LLaVA
export HF_HOME="/workspace"
python -m llava.serve.api -H 0.0.0.0 -p 5000

Sending requests to the Flask API

You can use the test script to test your API.

Acknowledgements

1. Matthew Berman for giving me a demo on LLaVA, as well as his amazing YouTube videos.

Community and Contributing

Pull requests and issues on GitHub are welcome. Bug fixes and new features are encouraged.

Appreciate my work?