common_fit_params: failed to fit params to free device memory — n_gpu_layers set to 999999, abort (AMD RX 9060 XT)

[titansonico10]

Which version of LM Studio?

LM Studio 0.4.18

Which operating system?

Windows 11

What is the bug?

When loading a model, LM Studio logs a warning from common_fit_params indicating it failed to fit parameters to free device memory because n_gpu_layers was already set by the user to 999999, causing an abort.

Model

gemma4-12b

GPU

AMD RX 9060 XT — 16 GB VRAM

Logs

2026-07-04 00:23:03 [DEBUG]
 0.03.913.323 W common_fit_params: failed to fit params to free device memory: n_gpu_layers already set by user to 999999, abort

Expected behavior

LM Studio should either adjust n_gpu_layers automatically to fit available VRAM, or show a clear user-facing error instead of silently aborting the param fitting process.

Additional context

The value 999999 for n_gpu_layers appears to be a sentinel/default value used by LM Studio to mean "offload all layers". When the automatic fitting logic detects this user-set value, it aborts instead of falling back gracefully.

[titansonico10]

Additional observation: while the logs show n_gpu_layers = 999999, the LM Studio UI displays 48 GPU layers — which is the actual layer count of Gemma 4 12B.

This suggests 999999 is a sentinel value used internally by LM Studio to mean "offload all layers", but this raw value is passed directly to the llama.cpp backend instead of being resolved to the real layer count first. The common_fit_params logic then encounters this sentinel value, treats it as a user-set override, and aborts the automatic parameter fitting instead of falling back gracefully.

Expected behavior: LM Studio should resolve the sentinel 999999 to the actual model layer count (48 in this case) before passing it to llama.cpp, or common_fit_params should handle this sentinel value correctly.