llmfan46/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-uncensored-heretic

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 30, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The llmfan46/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-uncensored-heretic is a 12 billion parameter Gemma 4 model, fine-tuned by llmfan46, that has been decensored using Heretic v1.4.0. This model significantly reduces refusals by 87% compared to its original counterpart while maintaining model quality with a KL divergence of 0.0367. It is specifically optimized for coding and agentic tasks, demonstrating a 3.5x higher score on the tau2-bench telecom agentic tool-use benchmark than the base model.

Loading preview...

Model Overview

This model, llmfan46/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-uncensored-heretic, is a 12 billion parameter Gemma 4 variant that has undergone a decensoring process using Heretic v1.4.0. It is based on yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF and aims to provide a less restrictive experience.

Key Capabilities

  • Reduced Refusals: Achieves an 87% reduction in refusals (13/100 vs 99/100 for the original) with a minimal KL divergence of 0.0367, indicating preserved quality.
  • Agentic and Coding Focus: Specifically fine-tuned for coding, terminal operations, and agentic workflows, including multi-step tool-use trajectories.
  • Enhanced Agentic Performance: Scores approximately 3.5x higher than the base model on the tau2-bench telecom agentic tool-use benchmark.
  • Grounded Reasoning: Exhibits 0% fabrication on coding/terminal fabrication probes, grounding its actions by first using commands like grep, read, or ls.

When to Use This Model

  • Agentic Applications: Ideal for tasks requiring tool use, debugging, and multi-step technical problem-solving.
  • Coding Tasks: Suited for code generation, understanding, and related development workflows.
  • Uncensored Use Cases: When a model with significantly fewer content refusals is required, particularly for creative or less constrained applications.

Limitations

  • General Knowledge Trade-off: As a focused fine-tune, it may perform slightly below the base model on general-knowledge benchmarks like MMLU-Pro.
  • English-centric: Primarily designed and optimized for English language tasks.

Finetunes
3 models