DeepSeek Model Exploration and Streamlit App

Overview

This repository provides a structured approach to exploring and utilizing the DeepSeek-R1-Distill-Qwen-1.5B model for text generation, problem-solving, and reasoning tasks. It includes a Jupyter Notebook for in-depth experimentation and a Streamlit app that offers a user-friendly interface to interact with the model.

This session was conducted by Attreyee Mukherjee and Piyush Chugeja, on behalf of CodeCell, on February 13, 2025, from 11 AM to 1 PM.

Prerequisites

Before running the notebook or Streamlit app, ensure you meet the following requirements:

• Access to DeepSeek on Hugging Face: Make sure you have permission to use the deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B model from Hugging Face.
• Together AI API Key: If using the Streamlit app, you need an API key from Together AI to interact with the model.
• Python Environment: Ensure you have Python installed (preferably Python 3.8+).
• Required Libraries: Install necessary dependencies such as transformers, torch, streamlit, and python-dotenv.

Notebook Details

The Jupyter Notebook demonstrates how to:

• Load the DeepSeek model and tokenizer using the Hugging Face Transformers library.
• Generate structured responses with thought reasoning and markdown-formatted output.
• Use structured prompts for solving mathematical problems, coding tasks, and complex reasoning challenges.

Streamlit App

The Streamlit application provides a simplified way to interact with the model:

• Users can select from different problem types: Math, Code, or Complex Reasoning.
• The app formats responses with structured <think> and <response> tags.
• Responses include markdown and LaTeX formatting for mathematical equations.
• The app fetches solutions from the Together AI API and presents structured answers.

Running the Streamlit App

To launch the app, simply run the Streamlit script and enter your problem statement. The model will process your input and generate a well-structured response.

Conclusion

This repository serves as a guide for exploring the DeepSeek-R1-Distill-Qwen-1.5B model in both research and practical applications. Whether through the Jupyter Notebook or the interactive Streamlit app, users can leverage DeepSeek’s capabilities for structured reasoning and problem-solving.