Vision model fails to load for Gemma-4 12B (unknown projector type: gemma4uv)

[skidu]

What happened?

When attempting to load the new Gemma-4 12B model along with its corresponding multimodal projector (mmproj), LM Studio crashes/fails to load the model. The text-only model loads perfectly fine, but the auto-detection of the vision projector throws an "unknown projector type: gemma4uv" error.

It seems the underlying llama.cpp in the current build (0.4.12) hasn't been updated to support the new gemma4uv encoder-free architecture introduced in the 12B version.

Steps to reproduce

1. Download a Gemma-4 12B text model (e.g., gemma-4-12B-it-heretic-Q4_K_M.gguf).
2. Download the official Gemma-4 12B vision projector (mmproj-gemma-4-12B-it-f16.gguf from bartowski).
3. Place both files in the same directory so LM Studio auto-loads the vision projector.
4. Try to load the model. It fails with a CLIP init error.

Hardware / Software

• OS: Windows 11
• LM Studio Version: 0.4.12 (build 1)
• CPU/GPU: AMD Ryzen AI 9 HX 370 / Radeon 890M
• RAM: 64GB

Relevant Log Output

clip_model_loader: has vision encoder
clip_model_loader: has audio encoder
clip_ctx: CLIP using Vulkan0 backend
clip_init: failed to load model 'C:/.../mmproj-gemma-4-12B-it-f16.gguf': load_hparams: unknown projector type: gemma4uv

mtmd_init_from_file: error: Failed to load CLIP model...
srv load_model: failed to load multimodal model...