Latest Official Google Gemma 4 GGUF Models Fail to Load in LM Studio 0.4.19

[hosiken]

Which version of LM Studio?
LM Studio 0.4.19 (Build 2)
llama.cpp Runtime 2.25.2 (CUDA 12 / CPU)

Which operating system?
Windows 11
GeForce RTX 4080 Laptop GPU

What is the bug?
The official GGUF files provided by Google for the updated version of Gemma 4, released on July 15, 2026, cannot be loaded.

The models fail to load regardless of whether the llama.cpp runtime is set to CUDA 12 or CPU.

The versions available before this update could be loaded and used normally without any issues.

Screenshots
None

Logs

Engine protocol runtime llama-server for c+rpmVpyTIi80FnZFIEiB5bN exited before becoming healthy. exitCode=3221226505, signal=null

To Reproduce
Steps to reproduce the behavior:

1. Download any of the following models and place it in the model folder. The issue occurs regardless of whether an mmproj file is present.

   • https://hf.co/google/gemma-4-12B-it-qat-q4_0-gguf/
   • https://hf.co/google/gemma-4-E2B-it-qat-q4_0-gguf/
   • https://hf.co/google/gemma-4-E4B-it-qat-q4_0-gguf/
   • https://hf.co/google/gemma-4-26B-A4B-it-qat-q4_0-gguf/
   • https://hf.co/google/gemma-4-31B-it-qat-q4_0-gguf/

2. Load the model from the model selector in LM Studio.

3. Observe that the error shown in the log above occurs.

[TomerGamerTV]

I'm getting this error when trying to load it from the bionic app:

Engine protocol ngPredictTokens request returned 400: {"error":{"code":400,"message":"Assistant message must contain either 'content' or 'tool_calls'!","type":"invalid_request_error"}}

[hosiken]

Google has since updated their official GGUF files again, and after updating to the latest version, the error no longer occurs.

Since the issue was not caused by LM Studio, I am closing this issue.