Releases: ashleykleynhans/llava-docker
Releases · ashleykleynhans/llava-docker
1.7.0
- Bumped CUDA to version 12.1.
- Bumped base docker image version to 1.7.0.
- Added code-server on port 7777.
- Keep venv on container disk instead of syncing to /workspace to significantly reduce start up time with RunPod Network Volumes.
1.6.0
- Bumped to latest LLaVA commit.
- Added
JUPYTER_LAB_PASSWORD
so that a password can be set to access Jupyter Lab. - Use new base image to simplify maintaining and building the image.
- Added badges.
1.4.6
- Use buildx bake to build the Docker image.
- Bump runpodctl to version v1.14.2.
- Add xformers 0.0.23.post1 back.
- Fixed bug where SSH service didn't start if no
PUBLIC_KEY
environment variable was set.
1.4.5
- Removed password for Jupyter.
- Added environment variable
VENV_PATH
to enable setting a custom path for the application venv, with a default of/workspace/venvs/llava
. - Supports sharing a network volume with other applications.
1.4.4
- Added OhMyRunPod.
- Added RunPod File Uploader.
- Bumped runpodctl to version 1.13.0.
1.4.3
- Bumped to latest LLaVA commit.
- Fixed Flask API.
1.4.1
- Bumped to latest LLaVA commit.
- Prevent
start.sh
script from crashing on errors.
1.4.0
- Bumped to latest LLaVA commit.
- Bumped torch to version 2.1.2.
- Removed xformers.
- Download the default model (
liuhaotian/llava-v1.6-mistral-7b
) into the Docker image so that it does not have to be downloaded at runtime. - Sync the model to /workspace when the container starts so that models can persist on persistent storage.
- Start ssh and Jupyter before syncing to /workspace.
1.3.2
- Bumped to latest LLaVA commit.
- Reverted
ENTRYPOINT
toCMD
so that docker command overrides can be applied.