This is a backup quant inferior to mradermacher/Gemma-4-31B-Cognitive-Unshackled-i1-GGUF. I recommend: Gemma-4-31B-Cognitive-Unshackled.i1-IQ4_XS.gguf as a replacement.
Okay. So I ran this in llama.cpp and Silly Tavern chat completion.
On my 24VRAM this fits 32000 context at f16 kv cache and 1024 batch.
Works fine, long RP context coherence, thinking and no thinking. No breaking, no issues.
