Qwen2.5-Math-1.5B Local Machine Implementation

This project demonstrates how to run the Qwen2.5-Math-1.5B model locally for mathematical reasoning and language generation using Hugging Face Transformers.

Requirements

• Python 3.8+
• PyTorch (with GPU support for better performance)
• Transformers
• Accelerate (optional, for device mapping)

Install dependencies:

 pip install torch transformers accelerate

How to Run

1. Download the model locally:

Download the model files from Hugging Face and place them in a directory, e.g., /path/to/Qwen2.5-Math-1.5B.

2. Edit the notebook:

Update the path to the downloaded model:

tokenizer = AutoTokenizer.from_pretrained('/path/to/Qwen2.5-Math-1.5B')
model = AutoModelForCausalLM.from_pretrained('/path/to/Qwen2.5-Math-1.5B', torch_dtype=torch.bfloat16)

3. Run the notebook: Launch Jupyter Notebook and execute each cell to:

• Tokenize an input prompt

• Perform inference

• Decode the model output

Example:

input_text = "Solve: 123 + 456 = ?"
inputs = tokenizer(input_text, return_tensors='pt')
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0]))

Model Details

• Model: Qwen2.5-Math-1.5B

• Type: Causal Language Model

• Use Case: Advanced reasoning and mathematical problem solving

• Framework: Hugging Face Transformers

Notes

• Make sure your device supports bfloat16 or change torch_dtype=torch.float16 or float32 as needed.

• Consider using model.to("cuda") for faster inference on GPU.

Reference

• Qwen2.5-Math-1.5B on Hugging Face

• Transformers Documentation