🚨⚠️ I HAVE REACHED HUGGING FACE'S FREE STORAGE LIMIT ⚠️🚨
I can no longer upload new models unless I can cover the cost of additional storage. I host 70+ free models as an independent contributor and this work is unpaid. Without your support, no more new models can be uploaded.
Your help will motivate me and would go into further improving my workflow and coverings fees for storage, compute and may even help uncensoring bigger model with rental Cloud GPUs.
Lower refusals indicate fewer content restrictions, while lower KL divergence indicates more closeness to the original model's baseline. Higher refusals cause more rejections, objections, pushbacks, lecturing, censorship, softening and deflections.
MMLU - Massive Multitask Language Understanding, multiple-choice questions across 57 subjects (math, history, law, medicine, etc.).
Quantizations
For the K-quants below, selected Gemma 4 attention and FFN tensors are kept at higher precision where useful.
Gemma 4 does not use the ssm_alpha, ssm_beta, or ssm_out tensors found in some Qwen-style hybrid/SSM architectures. Instead, these GGUFs preserve key Gemma 4 attention projection tensors at higher precision.
Q6_K uses a higher-quality XL-style layout:
attn_q, attn_k, attn_v, and attn_output are kept as Q8_0.
ffn_gate, ffn_up, and ffn_down are kept as Q8_0.
ffn_down_exps is requested as Q6_K where supported. Some tensors may fall back to Q8_0 due to Gemma 4 tensor shape constraints.
Q5_K_M, Q5_K_S, Q4_K_M, and Q4_K_S keep the main attention projection tensors as Q8_0:
attn_q
attn_k
attn_v
attn_output
Q3_K_L and Q3_K_M keep the main attention projection tensors as BF16:
attn_q
attn_k
attn_v
attn_output
This helps preserve Gemma 4’s attention path at higher precision, especially for lower-bit quants, while avoiding large file-size increases from unnecessarily up-quantizing the largest MoE expert tensors.
A Vision Projector File is Required for vision/multimodal capabilities. Use alongside any quantization above.
Usage
Works with llama.cpp, LM Studio, Ollama, and other GGUF-compatible tools.
image
Mero Mero
Gemma4 26B A4B
01Overview
God, this model was difficult to work with.
Google cooked, there wasn't a lot to improve but there was a lot to break.
This model is a finetune that was merged back into the original instruct. It feels a lot like the original instruct. However, reasoning is more structured, using less tokens during RP and this model generally has a slightly less verbose / flowery writing style.
Main weakness of this model I think is the swipe variety hasn't improved. Logic and repetition I think are roughly on par with the original.
Despite using 35 million tokens, this dataset is fairly modest in size. Trainable is somewhere in the rough ballpark of 15 million. The extra tokens are from a new multi turn RP dataset that I train last turn only.
Feels like Google left the instruct model at the razor's edge of overfitting. Finetune it at all and it feels like it'll rapidly lose intelligence, despite taking the writing style nicely. Hard to tell if you're overfitting or underfitting.
My solution was to blast the model with my data anyway to ensure it picked up the new reasoning format and writing style and then merge that back into the instruct to heal the logic damage. There's still room for a better merge that keeps more of the writing style and potentially using the base model to undo some of the overfitting.
