reedmayhew/Grok-3-gemma3-4B-distilled

xAI Grok 3

Distilled - Gemma 3 4B

NEW REASONING VERSION AVAILABLE:

https://huggingface.co/reedmayhew/Grok-3-reasoning-gemma3-4B-distilled-GGUF

Overview

This model is a Gemma 3 4B variant distilled from xAI’s Grok 3. It was fine-tuned to emulate Grok’s depth and structured clarity, particularly in tasks involving complex thought, such as problem-solving, coding, and mathematics.

Technical Details

• Developed by: reedmayhew
• Base Model: google/gemma-3-4b-it
• Training Speed Enhancement: Trained 2x faster with Unsloth and Huggingface's TRL library

Training Data

The model was trained on:

• reedmayhew/Grok-3-100x

This dataset consists of 100 high-quality Grok 3 completions responding to deep questions, solving math problems, and writing or analyzing code. The aim was to distill Grok’s analytical approach and technical versatility into a smaller, accessible model.

This Gemma model was trained 2x faster with Unsloth and Huggingface's TRL library.