tpls/gemma-4-12B-coder-fable5-composer2.5-v1-abliterated

Overview

This model, tpls/gemma-4-12B-coder-fable5-composer2.5-v1-abliterated, is a 12 billion parameter variant of the Gemma-4 Coder series. It is provided as uncensored weights (safetensors) for fine-tuning, merging, or quantizing. The model has been processed using an 'abliteration' technique, which removes refusal directions from its weights, making it capable of attempting requests that a standard model would typically refuse.

Key Characteristics

• Uncensored: Safety guardrails have been substantially removed, allowing it to generate content a stock model would refuse.
• Tool-calling: Designed for agentic tool use, though its native tool markup requires a recovery shim (e.g., tpls/gemma4-tool-shim) for standard tool_calls parsing with llama.cpp --jinja.
• Deprecated: This specific version is deprecated and superseded by tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v5-abliterated, which applies the same uncensoring technique to a newer, tool-calling optimized base.

Intended Use

This model is built for:

• Code generation.
• Agentic tool use.
• Serving locally via llama.cpp / Ollama (GGUF quantizations are available separately).
• As a base for further fine-tuning, merging, or quantizing.

Users are responsible for validating outputs and managing the implications of its uncensored nature.