Disable thinking of Gemma-4-E4B and Gemma-4-E2B on LM Studio? Thinking-button does not stop thinking, just does not hide it inside "thinking" block?
Posted by film_man_84@reddit | LocalLLaMA | View on Reddit | 7 comments
So as the title says, I try to disable thinking on Gemma 4 on models E2B and E4B in LM Studio.
When I press "Think"-button to disable it, it will visually seems to disable it but does not disable it from responses. It shows thinking patterns on the chat anyway but those does not go anymore under "Thinking" block what can be hidden, instead it just echos whole thinking process to chat?
I tried to edit Jinja template but without success.
Note that I don't have this issue with bigger models - disabling thinking works as excepted. Have any of you any success with this on smaller models?
Altruistic-Theme432@reddit
The Jinja template used by LM Studio has a bug. You can copy a correct template from elsewhere, overwrite the LMS template, and then you can correctly enable or disable Thinking.
Take this as an example, copy the content in "tokenizer.chat_template".
https://huggingface.co/unsloth/gemma-4-E4B-it-GGUF/blob/main/gemma-4-E4B-it-UD-Q6_K_XL.gguf
parronym@reddit
I noticed the exact same issue when I tried to disable thinking on the Gemma-4-E4B in LM Studio. Thinking still happens, but it now happens directly in the output together with the actual output. Hopefully it gets fixed soon.
kal_0008@reddit
seems impossible to stop it, even when I try it in Ollama. This model wants to always think no matter what
TemporaryUser10@reddit
Do you chat with it through the LM Studio chat interface, or through a third party connected to LM Studio?
film_man_84@reddit (OP)
Through LM Studio chat interface. When I tried this with Koboldcpp + SillyTavern I had no same issue, so it is related to LM Studio chat interface only.
FamousFlight7149@reddit
You should download the official GGUF from the LM Studio site, they always work perfectly for this. You shouldn’t download GGUFs that aren’t marked as ‘Staff picks’ unless you can create the manifest and yaml files yourself. https://lmstudio.ai/models/gemma-4
FamousFlight7149@reddit
You should download the official GGUF from the LM Studio site, they always work perfectly for this. You shouldn’t download GGUFs that aren’t marked as ‘staff pick’ unless you can create the manifest and yaml files yourself. https://lmstudio.ai/models/gemma-4