Incorrect output during inference
#1
by Ike - opened
I'm trying to use gemma-4-26B-A4B-it-MXFP4_MOE_BF16.gguf from here but any prompt with any setting is showing this output:
own<unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49><unused49> (it doesn't stop)
Tested on LM Studio 0.4.9 (Build 1), llama.cpp v2.11.0 on Windows.
I also tried to compile llama.cpp b8662 on linux and the results are about the same
It seems they're still fixing some issues in llama.cpp.
Just now they released b8665, that has specific fixes for gemma4.
This release for me works fine now.
noctrex changed discussion status to closed