1.3.0

• Bumped LLaVA to v1.2.0 (LLaVA 1.6).
• Changed default model to liuhaotian/llava-v1.6-mistral-7b.
• Changed Flask API default model to liuhaotian/llava-v1.6-mistral-7b.

1.2.5

• Fixed a bug in the fix_venv.sh script that was causing the venvs not to be fixed on a RunPod Network Volume.
• Updated rsync commands.
• Updated the HTTP 502 error page to provide additional information.
• Refactored the pre_start.sh script to use functions.
• Only sync when necessary to speed up startup time.
• Create FileZilla config for Runpod at /workspace/filezilla_sftp_config.xml if the pod has a public IP.

1.2.4

• Bumped to latest LLaVA commit.
• Changed CMD to ENTRYPOINT to reduce logging verbosity.
• Fixed missing protobuf module which was causing issues loading the original LLaVA models.

1.2.3

• Updated the Flask API so that it does not load the model on every request - see Flask API docs.

1.2.1

• Added support for llava_mistral model types.
• Now defaults to the SkunkworksAI/BakLLaVA-1 model which is a llava_mistral model.

1.1.2

• Now defaults to the liuhaotian/llava-v1.5-7b model so that it can run on 24GB of VRAM instead of requring an A6000+.
• Added support for providing a MODEL environment so that a different model version (eg. liuhaotian/llava-v1.5-13b) can be used.

1.1.0

Initial release.