Model Card for richterdc/deepseek-coder-finetuned-tdd

This model is fine-tuned to help developers generate test cases from code or plain language descriptions. It is designed to support Test-Driven Development (TDD) by suggesting tests that can improve code quality.

Model Details

Model Description

This model has been fine-tuned to generate test cases for software code. It takes in code snippets or descriptions of functionality and suggests relevant tests. The model uses the Hugging Face Transformers library and is deployed as a Flask API. It is built for fast inference with GPU support and is intended to help developers by automating part of the TDD process.

Developed by: Richter Dela Cruz
Funded by [optional]: Angelo Richter L. Dela Cruz, Alyza Reynado, Gabriel Luis Bacosa, and Joseph Bryan Eusebio
Shared by [optional]: Angelo Richter L. Dela Cruz
Model type: Causal language model for code generation and understanding
Language(s) (NLP): English (for code and comments)
License: Not specified
Finetuned from model [optional]: Fine-tuned from a base code generation model (deepseek-ai/deepseek-coder-1.3b-instruct)

Model Sources [optional]

Repository: https://github.com/RichterDelaCruz/tdd-deployment
Paper [optional]: On going
Demo [optional]: See the API demo instructions in the repository

Uses

Direct Use

The model can be used to generate test cases directly from code snippets or textual descriptions. This is useful for developers who want to quickly get ideas for tests to cover their code.

Downstream Use [optional]

The model can also be integrated into larger development pipelines or fine-tuned further for specific applications. For example, it can be used within continuous integration systems to suggest tests for new code changes.

Out-of-Scope Use

This model is not designed for generating security-critical tests or for replacing thorough human testing. It may not capture all edge cases and should not be solely relied upon for complete test coverage.

Bias, Risks, and Limitations

The model may generate test cases that are too generic or miss specific edge cases.
It might produce plausible-looking tests that require manual review.
Its performance may vary depending on the complexity of the input code or description.

Recommendations

Users should always review the generated test cases before using them in production. Fine-tuning on domain-specific data is recommended to improve relevance and accuracy.

How to Get Started with the Model

Clone the Repository:

git clone https://github.com/RichterDelaCruz/tdd-deployment.git
cd tdd-deployment

Install Dependencies:
```
pip install -r requirements.txt
```
Run the Flask API:
```
python generate-test.py
```

Test the API:

Use curl or any API testing tool:

curl -X POST "http://localhost:8000/generate" \
  -H "Content-Type: application/json" \
  -d '{"input_text": "Write a Python function to add two numbers"}'

Training Details

Training Data

The exact details of the training data are not provided. It likely consists of publicly available code repositories and associated test cases.

Training Procedure

The model was fine-tuned using standard practices for causal language models on a dataset of code and test cases.

Preprocessing [optional]

Preprocessing steps were applied to prepare the code and test case data, though specific details are not provided.

Training Hyperparameters

Training regime: Standard fine-tuning for causal language models (e.g., using PyTorch with mixed precision)

Speeds, Sizes, Times [optional]

The model is optimized for GPU inference and has been tested on hardware such as the RTX 3090 for scalability.

Evaluation

Testing Data, Factors & Metrics

Testing Data

Details about the testing dataset are not provided. Evaluation likely used code examples and corresponding expected test cases.

Factors

Evaluations may consider code complexity, coverage, and the correctness of the generated tests.

Metrics

Metrics might include improvements in test coverage or the accuracy of the suggested test cases, though specific metrics are not documented.

Results

Evaluation results are not comprehensively documented. Users are encouraged to evaluate the model based on their own codebases.

Summary

The model is effective at generating plausible test cases for a variety of code snippets, though manual review is recommended to ensure correctness and completeness.

Model Examination [optional]

No detailed interpretability or analysis work has been provided for this model.

Environmental Impact

Carbon emissions for model training and inference can be estimated using the Machine Learning Impact calculator.

Hardware Type: RTX 3090 or similar GPU
Hours used: Varies by deployment
Cloud Provider: Vast.ai (or any provider with CUDA support)
Compute Region: Not specified
Carbon Emitted: Not specified (estimate using the ML Impact calculator)

Technical Specifications [optional]

Model Architecture and Objective

This model is based on a causal language model architecture, fine-tuned specifically for code generation and test case creation. Its objective is to assist developers in following Test-Driven Development practices.

Compute Infrastructure

The model is deployed as a Flask API using gunicorn for scalability, with PyTorch handling model inference.

Hardware

The model runs on both CPU and GPU, with best performance observed on GPUs.

Software

Built using Python, Flask, PyTorch, and Hugging Face Transformers.

Citation [optional]

BibTeX:

@misc{richterdc2025tdd,
  author = {Richter Dela Cruz},
  title = {richterdc/deepseek-coder-finetuned-tdd},
  year = {2025},
  publisher = {GitHub},
  howpublished = {\url{https://github.com/RichterDelaCruz/tdd-deployment}}
}

APA:

Richter Dela Cruz. (2025). richterdc/deepseek-coder-finetuned-tdd. GitHub. Retrieved from https://github.com/RichterDelaCruz/tdd-deployment

Glossary [optional]

Test-Driven Development (TDD): A development approach where tests are written before the code to ensure functionality.
Flask: A Python web framework used for building APIs.
Transformers: A library by Hugging Face for working with state-of-the-art language models.

More Information [optional]

For further details, visit the repository.

Model Card Authors [optional]

Angelo Richter L. Dela Cruz, Alyza Reynado, Gabriel Luis Bacosa, and Joseph Bryan Eusebio

Model Card Contact

For inquiries or contributions, please reach out via GitHub.

richterdc
/

deepseek-coder-finetuned-tdd