How can we bring VLM of choice?

we have choice of VITA and GPT,
How can I use different endpoint from maybe build.nvidia.com or any other openAI compatible model

If you have another openAI compatible model you can run the following:

docker run … -e OPENAI_API_KEY=<OPENAI_API_KEY>
-e VIA_VLM_OPENAI_MODEL_DEPLOYMENT_NAME=gpt-4o
-e VLM_MODEL_TO_USE=openai-compat

You can also load a custom model:

  1. Mount the directory containing the inference.py file and the optional model files in the
    container.
  2. Set the MODEL_PATH env variable to the mount path. MODEL_PATH should point to the
    directory containing the inference.py file in the container.
  3. Set VLM_MODEL_TO_USE=custom

ls <MODEL_DIR_ON_HOST>
inference.py
docker run … -v <MODEL_DIR_ON_HOST>:<MODEL_DIR_IN_CONTAINER>
-e MODEL_PATH=<MODEL_DIR_IN_CONTAINER> -e VLM_MODEL_TO_USE=custom …

You can refer to pages 21 and 23 of the VIA DP User Guide for more details.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.