tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v3

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 24, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

The tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v3 is a 12 billion parameter Gemma-4 based model, specifically designed for code generation and agentic tool use. This model provides GGUF quantizations for local serving via llama.cpp or Ollama, supporting native Jinja tool-calling. It is an intermediate SFT iteration, now deprecated and superseded by v5, but was built to facilitate local development and fine-tuning for coding tasks.

Loading preview...

Model Overview

This model, tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v3, is a 12 billion parameter Gemma-4 based language model. It is specifically fine-tuned for code generation and agentic tool use, providing GGUF quantizations optimized for local deployment with llama.cpp or Ollama. A key feature is its native support for Jinja-based tool-calling, enabling more structured interactions for agentic workflows.

Key Characteristics

  • Architecture: Based on the Gemma-4 12B Coder family.
  • Parameter Count: 12 billion parameters.
  • Context Length: Supports a context length of 32768 tokens.
  • Deployment: Designed for local serving using llama.cpp or Ollama, with recommended Q4_K_M quantization for a balance of size and quality, fitting within 12 GB VRAM.
  • Tool-calling: Features native --jinja support for agentic tool use, facilitating structured output for function calls.

Important Note

This version is deprecated and has been superseded by tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v5-GGUF. Users are advised to use the newer v5 iteration for new projects. While it was an intermediate SFT iteration, it served as a foundation for further development in the Gemma-4 Coder family.

Finetunes
1 models