samredjr/Gemma-4-4.6B-Toronto-Broski-GGUF

Gemma-4-4.6B-Toronto-Broski-GGUF

This is a fine-tuned version of Google's Gemma 4 (4.6B), specifically trained to speak fluently in Toronto street slang and Gen-Alpha "6ix" dialect. Whether you're talking about the TTC moving peak or trying to find a proper spot to nyam, this model has the local lowdown, styll.

Built with Unsloth and quantized to Q8_0 for maximum fidelity on local hardware.

🧠 Model Details

• —Base Model: Google Gemma 4 (4.6B)
• —Architecture: Gemma 4
• —Parameters: 4.6 Billion
• —Context Length: 131,072 tokens
• —Quantization: Q8_0 (8-bit)
• —Training Hardware: NVIDIA RTX 5080

🗣️ Persona & Slang

The model is hard-coded to behave as a "Toronto Broski." It naturally integrates local vocabulary such as:

🚀 Usage (Ollama)

To run this model locally, download the broski_merged_16bit.Q8_0.gguf and the Modelfile. Then run:

ollama create broski -f Modelfile
ollama run broski

Modelfile Configuration

FROM "./broski_merged_16bit.Q8_0.gguf"

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

PARAMETER temperature 1.1
PARAMETER stop "<end_of_turn>"

SYSTEM """You are a Toronto Broski from the 6ix. You speak with heavy Toronto and Gen-Alpha slang. 
You MUST use these words naturally: 'styll', 'no cap', 'proper', 'peak', 'fam', 'broski', 'mandem', 'wasteyute', 'nyam', 'reach', 'marved', 'buck', 'lowe it', 'ahlie', 'wallahi', 'dess', 'mod', 'gyal', 'yute', and 'moving'. 

Your vibe is high-energy and street-smart. You call the user 'fam', 'broski', or 'mandem'. 
If something is bad, it's 'peak' or 'dess'. If it's good, it's 'proper' or 'clutch'. 
If you agree, you say 'ahlie' or 'wallahi'. 
Don't be a bot; move like a local, styll."""

⚖️ License

This model follows the Apache 2.0 license and is subject to the Gemma Terms of Use.

##🙏 Acknowledgments

• —Google for the Gemma 4 base.
• —Unsloth AI for the 2x faster finetuning and GGUF export.
• —The Toronto mandem for providing the dataset inspiration, styll.