gemma 4 A4B 98e v6 coder it GGUF ✅ Chat template — agentic loop fix (2026 06 10) The chat template inherited from Gemma 4 upstream re injected each prior tool call turn's reasoning content as a thought block (for assistant tool call steps after the last user message). In deep multi turn agentic / tool calling use (OpenCode, llama.cpp jinja , …) this fed the model an accumulating echo of its own reasoning and could trigger verbatim repetition loops . Single turn chat is unaffected. Validated on Q6 K: agentic loop rate 33% → 0% , with HumanEval+ 92.07% and MultiPL E 100 0.66 unchanged (the fix is single turn inert). A drop in chat template.fixed.jinja and a 1 minute template loop unittest.py are provided for audit. The bug is upstream wide across the Gemma 4 family + third party requants — see the writeup. As of 2026 06 10 the corrected template is embedded in every GGUF tier in this repo. The anti repetition sampler flags ( min p 0.05 repeat penalty 1.1 ) noted under How to Use are no longer required to prevent this agentic loop; they remain safe as general sampling. reasoning budget 8192 is still required for Gemma 4 thinking. GGUF quantizations of ManniX ITA/gemma 4 A4B 98e v6 cod…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy