devopsforflops/delia-functiongemma-270m-gguf

Delia FunctionGemma 270M - GGUF

This is the GGUF version of delia-functiongemma-270m, fine-tuned for Delia MCP tool orchestration.

Quick Start with Ollama

# Download the GGUF file
wget https://huggingface.co/devopsforflops/delia-functiongemma-270m-gguf/resolve/main/functiongemma-delia-f16.gguf

# Create Modelfile
cat > Modelfile << 'MODELFILE'
FROM ./functiongemma-delia-f16.gguf

TEMPLATE """{{ if .System }}<start_of_turn>developer
{{ .System }}
<end_of_turn>
{{ end }}<start_of_turn>user
{{ .Prompt }}
<end_of_turn>
<start_of_turn>model
"""

PARAMETER stop <end_of_turn>
PARAMETER stop <start_of_turn>
PARAMETER temperature 0.1
PARAMETER num_ctx 2048
MODELFILE

# Import to Ollama
ollama create functiongemma-delia -f Modelfile

# Test it
ollama run functiongemma-delia "Hello!"

Model Details

Property	Value
Base Model	google/functiongemma-270m-it
Architecture	Gemma3
Parameters	268M
Quantization	F16 (full precision)
File Size	~518 MB
Context Length	2048 tokens

Training

Fine-tuned using LoRA on Delia MCP tool calling examples:

• LoRA rank: 16
• LoRA alpha: 64
• Epochs: 20
• Dataset: 27 training examples from Delia test suite

Use with Delia

Add to your Delia settings.json:

{
  "model_dispatcher": {
    "name": "functiongemma-delia",
    "num_ctx": 2048
  }
}

Important: The model name must contain "functiongemma" for Delia to apply the correct prompt formatting.

Related Models

• delia-functiongemma-270m - Full merged HuggingFace model
• delia-functiongemma-270m-lora - LoRA adapter only

License

Apache 2.0