Experimental "Preserve Thinking" Jinja Template for Gemma4 31B in llama.cpp

<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1sic351/gemma_4_template_fix_channel_thought_leakage/"> <img alt="Gemma 4 template fix &lt;|channel&gt; /

<!-- SC_OFF --><div class="md"><p>I want to run only text prompts to Gemma4 with llama.cpp, but I dont want to use CLI or server - I want to have it fully embeded inside my code.

<!-- SC_OFF --><div class="md"><p>Beginner here, now I have a good GPU and ollama using docker. Pulled the Gemma4 weights and was able to add it to cursor using ngrok. </p>

<!-- SC_OFF --><div class="md"><blockquote> <p><strong>WARNING: wait before download from HF: I just realised my upload of the new versions with the additional fix in the chat

<!-- SC_OFF --><div class="md"><p>I like to be able to enable and disable thinking using a system prompt, so that I can control what which prompts generate thinking tokens rather

<!-- SC_OFF --><div class="md"><p>I've got the model Qwen3.6-35B-A3B-4bit running in oMLX, and I want to enable the kwarg preserve_thinking as described here: <a