Skip to content

Repository files navigation

Heat Equation LLM Challenge

Methodology

  1. Data Generation: Solved the heat equation using the Finite Difference Method (FDM) for Cases 1-4.
  2. Data Preprocessing: Extracted temperature distributions from VTK files and saved them in a structured CSV format.
  3. Fine-Tuning: Fine-tuned the IBM Granite 3.1 LLM using LoRA for parameter-efficient training.
  4. Evaluation: Generated responses to the questions in Questions.csv and created the submission.csv file.

Results

  • The fine-tuned model provided meaningful physical interpretations of the heat equation solutions.
  • Responses were evaluated using cosine similarity against ground truth answers.

Files

  • heat_equation_llm.ipynb: Kaggle notebook with the full implementation.
  • submission.csv: Submission file containing the model's responses.
  • results_lora/: Directory containing fine-tuning results.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages