💎 Gemma 3 27B IT Abliterated

This is an experimental, uncensored version of google/gemma-3-27b-it created using a new “abliteration” technique. It removes refusals while keeping most of the model’s capabilities intact.

Key Features: • Tool Calling Enabled: This version includes tool support, making it more flexible for various tasks. • Minimal Fine-tuning: The model has been minimally fine-tuned, focusing on improving output coherence and handling requests effectively.

Recommended Generation Parameters: • temperature=1.0 • top_k=64 • top_p=0.95

Abliteration - The model is abliterated by computing refusal directions based on hidden states for each layer independently.

Downloads last month
8
Safetensors
Model size
27.4B params
Tensor type
BF16
·
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.

Model tree for ZySec-AI/gemma-3-27b-tools

Finetuned
(10)
this model