Incorrect output in Gemma 4: seeking a solution to the problem
Hello! Could you please help me figure out how to solve an output issue with Gemma 4?
Any Gemma 4 — regardless of the quantization and version — produces characteristic artifacts (they vary, but these appear most frequently):
’’’’’’’’’’’’’’’’’’’’’’’’’’’’’’’’’
lilt’S a lilt’s a lilt…
a lS… de lS… l l S…
la la la a la l l l l l l l l l
lS lS lL lL lLL
one-size–//-//-
Backend: Oobabooga text generation WebUI (v4.4, 9dcf574, with Gemma 4 support);
SillyTavern (1.16.0 ‘release’ (e41bcf0cc)).
I’ve looked through the settings, disabled speculative decoding, and tried different templates — but to no avail.
Could the issue be related to decoding? How can this be fixed? Has anyone else encountered this problem? Or perhaps this error is recognizable — could someone suggest a way to deal with it?
I’d like to apologize in advance for my English — it’s not my native language.
Llamacpp may be out of date ; there have been a lot of updates to GEmma 4.
Likewise Google just re-issued jinja templates for the Gemma 4 in the past day or so.
You may need a newer quant / re-quant and/or llamacpp update(s).
Thank you for the response! The latest version of llama.cpp is included in release 4.4. But in any case, since others are also experiencing the errors and fixes/updates are still being released, this gives me hope. I’ll wait and try updating again.