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

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 23, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

This is a 12 billion parameter Gemma-4 Coder model from tpls, specifically the first supervised fine-tuning (SFT) iteration. It is designed for code generation and agentic tool use, offering native Jinja tool-calling capabilities. This version is deprecated and has been superseded by a newer iteration, primarily serving as weights for further fine-tuning, merging, or quantization.

Loading preview...

Overview

This model, tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft, is a 12 billion parameter Gemma-4 Coder model. It represents the initial supervised fine-tuning (SFT) iteration, providing model weights in safetensors format. It is built upon the yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1 base model.

Key Characteristics

  • Type: Model weights (safetensors) intended for fine-tuning, merging, or quantization.
  • Tool-calling: Features native Jinja (--jinja) support for agentic tool use.
  • Status: This specific version is deprecated and has been superseded by tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v5.

Intended Use & Limitations

This model is primarily built for code generation and agentic tool use. While it can be served locally via llama.cpp or Ollama, its main purpose is to serve as a base for further development, such as fine-tuning or merging. Users should be aware that outputs can be incorrect or fabricated, necessitating validation of tool arguments before execution and maintaining human oversight for critical applications. Ready-to-serve GGUF quantizations are available at tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-GGUF.