gemini-cli - 💡(How to fix) Fix Build program failure: no matching function for call to 'convert_float' on RTX 5080 with gpu_artisan backend [1 comments, 2 participants]

[Routing] GemmaClassifierStrategy failed: _ApiError: {"error":{"message":"failed to create the engine: failed to create engine for model │ │ gemma3-1b-gpu-custom with backend gpu_artisan: failed to create engine\n","code":500,"status":"Internal Server Error"}}

TL;DR

The Gemini CLI fails to initialize the local inference engine for the gemma3-1b-gpu-custom model due to an issue with FP16 calculations during the OpenCL kernel build process, resulting in the router always choosing Flash.

Guidance

• Verify that the GPU drivers are up to date and compatible with the Gemini CLI, as outdated drivers may cause issues with OpenCL kernel builds.
• Check the settings.json file to ensure that gemmaModelRouter is enabled, as this is a required setting for the Gemini CLI to function correctly.
• Attempt to disable FP16 calculations or switch to a different model that does not rely on FP16 calculations to see if the issue persists.
• Review the Gemini CLI documentation to see if there are any specific requirements or recommendations for running the gemma3-1b-gpu-custom model on the current system configuration.

Notes

The issue may be specific to the gemma3-1b-gpu-custom model or the system configuration, and further troubleshooting may be required to determine the root cause.

Recommendation

Apply a workaround by disabling FP16 calculations or switching to a different model, as this may allow the Gemini CLI to function correctly until a permanent fix is available.