Overview

TeichAI/Qwen3-4B-Instruct-2507-Gemini-3-Pro-Preview-Distill is a 4 billion parameter instruction-tuned model based on the Qwen3 architecture. Developed by TeichAI, this model was fine-tuned using a specialized Gemini 3 Pro Preview dataset. A unique aspect of its training involved removing the reasoning summaries from the dataset, with the model being fine-tuned exclusively on the final answers. This approach aims to distill direct answer generation capabilities from a dataset originally rich in reasoning effort.

Key Capabilities

• Specialized Training: Fine-tuned on a Gemini 3 Pro Preview dataset, focusing on direct answers.
• Efficient Training: Utilized Unsloth and Huggingface's TRL library for 2x faster training.
• Base Model: Built upon unsloth/Qwen3-4B-Instruct-2507.

Good For

• Coding: Designed to perform well in programming-related tasks.
• Science: Suitable for applications within scientific domains.