Gemma-4 E4B Gemini 3.1 Pro Reasoning Distill - GGUF

GGUF quantized versions of Ayodele01/gemma-4-E4B-Gemini-3.1-Pro-Reasoning-Distill.

Model Description

This is Google's Gemma-4 4B (E4B) instruction-tuned model, fine-tuned on Gemini 3.1 Pro reasoning datasets to improve chain-of-thought reasoning capabilities.

Training Data

• Roman1111111/gemini-3.1-pro-hard-high-reasoning
• Roman1111111/gemini-3-pro-10000x-hard-high-reasoning

Training Configuration

• Base Model: google/gemma-4b-it (via unsloth/gemma-4-E4B-it)
• Method: LoRA fine-tuning
• LoRA Config: r=8, alpha=8, dropout=0.1
• Learning Rate: 5e-5
• Epochs: 0.5
• Framework: Unsloth + TRL

Available Quantizations

Usage with llama.cpp

# Download a quantized model
wget https://huggingface.co/Ayodele01/gemma-4-E4B-Gemini-3.1-Pro-Reasoning-Distill-GGUF/resolve/main/gemma-4-E4B-Gemini-3.1-Pro-Reasoning-Distill-Q8_0.gguf

# Run with llama.cpp
./llama-cli -m gemma-4-E4B-Gemini-3.1-Pro-Reasoning-Distill-Q8_0.gguf \
  -p "What is the sum of all prime numbers between 1 and 50?" \
  -n 512

Usage with Ollama

Create a Modelfile:

FROM ./gemma-4-E4B-Gemini-3.1-Pro-Reasoning-Distill-Q8_0.gguf

TEMPLATE """<bos><start_of_turn>user
{{ .Prompt }}<end_of_turn>
<start_of_turn>model
"""

PARAMETER stop "<end_of_turn>"
PARAMETER temperature 0.7

Then:

ollama create gemma4-reasoning -f Modelfile
ollama run gemma4-reasoning

Recommended Settings

• Temperature: 0.7
• Top-P: 0.9
• Context Length: 8192 (model supports up to 128K)
• Repeat Penalty: 1.1

License

This model is released under the Gemma License.

Related Models

• Original Fine-tuned Model
• E2B Version