huihui-ai/MicroThinker-3B-Preview-v2

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:3.2BQuant:BF16Ctx Length:32kPublished:Jan 14, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

MicroThinker-3B-Preview-v2 by huihui-ai is a 3.2 billion parameter language model, fine-tuned from MicroThinker-3B-Preview, with a 32768-token context length. This iteration focuses on enhancing AI reasoning capabilities, demonstrating improved performance over its predecessor. It was fine-tuned using the FineQwQ-142k dataset, making it suitable for tasks requiring advanced reasoning.

Loading preview...

MicroThinker-3B-Preview-v2 Overview

huihui-ai's MicroThinker-3B-Preview-v2 is a 3.2 billion parameter language model, representing an advancement over its base model, MicroThinker-3B-Preview. This version has been specifically fine-tuned to improve AI reasoning capabilities, leveraging a substantial context length of 32768 tokens.

Key Training Details

  • Base Model: Fine-tuned from huihui-ai/MicroThinker-3B-Preview.
  • Dataset: Utilized 142k samples from the FineQwQ-142k dataset.
  • Training Environment: Fine-tuning was performed on a single RTX 4090 GPU (24GB).
  • Methodology: Employed Supervised Fine-Tuning (SFT) with LoRA (rank 8, alpha 32) for efficient adaptation.
  • Quantization: Trained with 4-bit quantization, using bfloat16 for compute and storage.
  • Epochs: Trained for 1 epoch with a max_length of 21710 tokens during fine-tuning.

Intended Use Cases

This model is designed for applications requiring enhanced reasoning, building upon the capabilities of its predecessor. Its fine-tuning process and focus on reasoning suggest suitability for tasks that benefit from step-by-step thinking and logical deduction.