MicroThinker-3B-Preview: Enhanced Reasoning Model

MicroThinker-3B-Preview is a 3.2 billion parameter language model developed by huihui-ai, fine-tuned from the huihui-ai/Llama-3.2-3B-Instruct-abliterated base model. Its primary focus is to enhance AI reasoning capabilities, making it suitable for tasks that require logical processing and step-by-step thought.

Key Characteristics

• Base Model: Fine-tuned from Llama-3.2-3B-Instruct-abliterated.
• Parameter Count: 3.2 billion parameters.
• Context Length: Supports a context length of 32768 tokens.
• Training Data: Fine-tuned using 142k entries from the FineQwQ-142k dataset.
• Training Environment: The fine-tuning process was conducted on a single RTX 4090 GPU (24GB) using Supervised Fine-Tuning (SFT) with LoRA.
• Quantization: Utilizes 4-bit quantization during training (quant_bits 4).

Use Cases

This model is particularly well-suited for applications where improved reasoning and structured thinking are crucial. Its training methodology, emphasizing a "think step-by-step" system prompt, suggests an optimization for tasks requiring logical deduction and problem-solving.

Deployment

MicroThinker-3B-Preview can be easily deployed and used with Ollama, providing a straightforward way to integrate it into various applications.