Q-Rely: AI Friend · public · first seen
・Fixed a timeout issue where the app could take 10+ minutes to fail during AI model loading. The app no longer falls back to the slow CPU server after a GPU timeout, so failures are detected much faster.
・If the NVIDIA CUDA server crashes during startup, the app now retries once with reduced VRAM settings before giving up.
・Startup now detects a hung AI server sooner (no output for 3 minutes) instead of always waiting the full timeout.
・The timeout error dialog now shows which backends were attempted, making it easier to report issues.
・Added a config option (llama_backend) to force a specific backend: "auto", "cuda13", "cuda12", "vulkan", or "cpu". Edit config.json next to Q-Rely.exe if you want to skip automatic fallback.
Thank you for the reports. If you still see errors, please send us the server_startup_failed_.txt file from the debug folder
| Depot | New manifest | Previous manifest |
|---|---|---|
| 5132011 | 96613037659069968 | 8622190275650772824 |