tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v4
The tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v4 model is a 12 billion parameter variant from the Gemma-4 family, developed by tpls. This model is specifically designed for code generation and agentic tool use, leveraging native Jinja for tool-calling. It is provided as safetensors weights for fine-tuning, merging, or quantizing, with ready-to-serve GGUF quantizations also available.
Loading preview...
Model Overview
The tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v4 is a 12 billion parameter model from the Gemma-4 family, provided as safetensors weights. It is primarily engineered for code generation and agentic tool use, supporting native Jinja for tool-calling. This specific version is marked as deprecated due to a tool-calling regression, where a shift towards conversational answering impacted its intended functionality. It has been superseded by tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v5.
Key Capabilities
- Code Generation: Designed to generate code effectively.
- Agentic Tool Use: Built to facilitate agentic workflows through tool integration.
- Native Jinja Tool-calling: Supports tool-calling using the Jinja templating language.
- Weights for Customization: Provided as safetensors weights, making it suitable for further fine-tuning, merging, or quantization.
Intended Use & Limitations
This model is intended for code generation and agentic tool use, with recommendations for local serving via llama.cpp or Ollama. Users should be aware that outputs may be incorrect or fabricated, necessitating validation of tool arguments before execution and maintaining human oversight for critical applications. Due to its deprecated status, new projects should utilize the successor model, tpls/gemma-4-12B-coder-fable5-composer2.5-v1-sft-v5.