FAQ

1. How long after startup is the service ready to use?

Initial startup takes 2 to 5 minutes to load the model onto the GPU. View progress with the following command:

docker-compose logs -f

When you see Uvicorn running on http://0.0.0.0:8000, the service is ready.

2. How do I manually load the Docker image?

gunzip -c docker-image-v1.0.0.tar.gz | docker load

3. What should I do if there is insufficient VRAM?

Adjust the following parameters in the .env file in the model deployment directory to reduce VRAM usage:

GPU_MEMORY_UTILIZATION=0.85
MAX_MODEL_LEN=8192

4. How do I verify the integrity of downloaded files?

If extraction fails or the service behaves abnormally, you can verify file integrity:

  1. Download the checksum file.
wget https://terra-1-g.djicdn.com/84f990b0bbd145e6a3930de0c55d3b2b/ai-models/qwen3-vl-32b-instruct/v1.0.0/checksums.md5
  1. Verify the files.
md5sum -c checksums.md5