Qwen3.5-2B Fine-tuned

The Challenge

Create an optimized fine-tuned version of Qwen3.5-2B that delivers high-quality theory exam Question answer ability tuned to SPPU university format, while training faster and more efficiently than traditional methods.

Solution

I leveraged Unsloth, a cutting-edge fine-tuning framework that enables 2x faster training compared to standard methods. The model was fine-tuned on custom datasets to enhance its theory question-answering abilities and university domain knowledge.

Technical Details

• Base Model: Qwen/Qwen3.5-2B
• Training Framework: Unsloth (2x faster training)
• Parameters: 2B
• Quantization: Q8_0 (8-bit)
• Model Size: 2.01 GB
• License: Apache 2.0

Key Features

• Optimized for theory question-answer responses
• Supports GGUF format for efficient inference
• Compatible with Transformers library
• Hardware compatible with 8-bit quantization
• Ready for deployment with text-generation-inference

Results

The fine-tuned model achieves improved theory question-answer performance while being significantly more efficient to train. Available on HuggingFace and supporting multiple quantization formats for flexible deployment.