Gemma-4-Gembrain-31B-it-uncensored-heretic Is Out Now, a Merge of Multiple Gemma 4 31B it Finetunes Designed to Boost Logical and Lateral Thinking for Improved Adherence, Increased Swipe Variety and Enhanced Creative Prose, With KLD of 0.0186 and 13/100 Refusals!
Posted by LLMFan46@reddit | LocalLLaMA | View on Reddit | 6 comments
Provided in both Safetensors and GGUFs.
Safetensors: llmfan46/Gemma-4-Gembrain-31B-it-uncensored-heretic: https://huggingface.co/llmfan46/Gemma-4-Gembrain-31B-it-uncensored-heretic
GGUFs: llmfan46/Gemma-4-Gembrain-31B-it-uncensored-heretic-GGUF: https://huggingface.co/llmfan46/Gemma-4-Gembrain-31B-it-uncensored-heretic-GGUF
I can make also GPTQs and NVFP4s if anyone asks for them.
Find all my models here: HuggingFace-LLMFan46
The original author of this finetune is: Nimbz
pigeon57434@reddit
i dont ever trust these weird merged models
Borkato@reddit
Does it require thinking on to be good for rp?
eidrag@reddit
I don't see the point of it, when the merge already done with other finetune, which itself already done with Gemma-4-31B-it-heretic-ara
LLMFan46@reddit (OP)
Yeah that's the confusing part, the model is supposed to be a merge of 7 different models, one of which is trohrbaugh/gemma-4-31b-it-heretic-ara that is supposed to have 5/100 refusals and yet Nimbz/Gemma-4-Gembrain-31B has 99/100 refusals! Which is EXACTLY the same amount of refusals as the base vanilla model google/gemma-4-31B-it and not only that, look at this:
MMLU test results:
Nimbz/Gemma-4-Gembrain-31B:
============================================================
============================================================
MMLU test results:
google/gemma-4-31B-it:
============================================================
============================================================
A model that is the result of a merge of 7 models, 6 of them being finetunes, shouldn't have numbers that either align or almost align with the original base vanilla model.
eidrag@reddit
i'm really tempted to try now lol. And they even managed to spell zerofata name wrong
LLMFan46@reddit (OP)
Oh I didn't even notice that haha.