Known Issues#
Pipeline server exits with 2 GPU streams#
Symptoms:
When two GPU pipeline streams are started, the pipeline server exits from the container.
Hardware:
Issue observed on BMG-580 discrete GPU.
WebRTC connectivity issues#
Symptoms:
Black video, no stream, or connection failures in the dashboard.
Checks:
Verify
HOST_IPin.envis reachable from the browser client.Confirm firewall rules allow the configured ports.
No models in dropdown#
Symptoms:
Model list is empty in the UI.
Checks:
Ensure
ov_models/contains at least one model directory with OpenVINO IR files.If you downloaded models, re-run the stack so the service rescans.
Pipeline server unreachable#
Symptoms:
Starting a run fails; backend reports it cannot reach the pipeline server.
Checks:
Ensure the
dlstreamer-pipeline-serverservice is running.Verify
PIPELINE_SERVER_URL(defaults tohttp://dlstreamer-pipeline-server:8080).
Port conflicts#
If the dashboard or APIs are not reachable, check whether the ports are already in use and update the .env values (for example DASHBOARD_PORT).
Performance/throughput lower than expected#
Larger VLMs require more compute and memory; try a smaller model.
Reduce
max_tokens.Ensure hardware acceleration and drivers are installed if using GPU.
Gemma model not working in GPU#
Gemma model is not working on GPU. Only working on CPU.
Limited testing on EMT-S and EMT-D#
This release includes only limited testing on EMT‑S and EMT‑D, some behaviors may not yet be fully validated across all scenarios.