Uw U 14 B Math v0.2

Introduction

The UWU-14B-MATH-V0.2 model is a fine-tuned version of the Qwen 2.5-14B model, enhanced for text generation tasks. It was developed by qingy2024 using responses from the QwQ 32B Preview and the NuminaMathCoT dataset.

Architecture

This model is based on the unsloth/qwen2.5-14b-bnb-4bit architecture. It utilizes the standard ChatML template for its operations.

Training

• Base Model: Qwen 2.5-14B
• Fine-Tuning Dataset: A verified subset of NuminaMathCoT, using Qwen 2.5 3B Instruct as a judge.
• QLoRA Configuration:
  • Rank: 32
  • Rank Stabilization: Enabled
• Optimization Settings:
  • Batch Size: 8
  • Gradient Accumulation Steps: 2 (Effective Batch Size: 16)
  • Warm-Up Steps: 5
  • Weight Decay: 0.01
• Training Steps: 500
• Hardware: A100-80GB GPU

Guide: Running Locally

1. Setup Environment: Ensure you have Python installed along with necessary libraries such as PyTorch and Transformers.
2. Clone Repository: Download the model files from the repository.
3. Install Dependencies: Use pip to install required packages (transformers, safetensors, etc.).
4. Load Model: Use the Transformers library to load the model.
5. Inference: Run inference scripts to generate text.

For optimal performance, consider using cloud GPUs such as NVIDIA A100s available from providers like AWS, GCP, or Azure.

License

The UWU-14B-MATH-V0.2 model is licensed under the Apache-2.0 License, allowing for broad use and modification.