iQS-VLM: Chat with the VLM through Open WebUI

Besides the live camera demo, iQS-VLM has a chat mode. It runs the same model, powered by GenieX, behind Open WebUI, so you can chat with the model and send it your own images from a browser on your development machine.

Chat mode needs no camera and no monitor.

Chat with the VLM

1. 1
   Start the demo in chat mode on the platform:

   {{ cbChat.fname }} bash

   This runs Qwen3-VL-4B on the NPU. To chat with LFM2.5-VL-450M instead, add the runtime and the backend (NPU, CPU, GPU or HYBRID):

   {{ cbChatLfm.fname }} bash

   The first chat start downloads the Open WebUI image, which takes a while. Later starts use the downloaded image.
2. 2
   When the model is ready, the terminal prints the Open WebUI address, http://<device_ip>:3000. Open it in a browser on your development machine.

   

3. 3
   Open WebUI opens with the model already selected. Type a message, or attach an image with +, and start chatting!

To stop chat mode, press q or Ctrl+C in the terminal.

Technical Details

1. 1
   Chat mode starts the model and an OpenAI-compatible API server on port 8000, then starts Open WebUI on port 3000 as a separate container connected to it. Your chats and settings are kept between runs.
2. 2
   You can also call the API directly, for example to list the model that is running:

   {{ cbCurl.fname }} bash

   It implements GET /v1/models and POST /v1/chat/completions. For the request format, see the OpenAI Chat Completions API.

3. 3
   Logs are saved on the platform in /var/lib/iqs-vlm-demo/logs. To follow them live while the demo runs, open another terminal on the platform:

   {{ cbTail.fname }} bash

   Log file	What it shows
   gateway.log	API requests from Open WebUI and the responses
   geniex.stdout.log, geniex.stderr.log	Model loading and inference output from GenieX
   lifecycle.log	Startup and shutdown steps