> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ecopi.chat/llms.txt
> Use this file to discover all available pages before exploring further.

# Development

> Learn how to develop and customize InternTA

<Info>
  **Prerequisites**:

  * NVIDIA GPU with 8GB or more VRAM
  * Python environment
  * Git
</Info>

## Development Setup

### 1. Clone the Repository

```bash theme={null}
git clone https://github.com/kongfoo-ai/internTA
cd internTA
```

### 2. Install Dependencies

```bash theme={null}
pip install -r requirements.txt
```

## Model Development

### Data Generation

The first step in customizing InternTA is preparing the training data. We support two types of fine-tuning data:

* Direct Q\&A data
* Guided Q\&A data

<AccordionGroup>
  <Accordion icon="database" title="Data Preparation Process">
    1. Compile a question bank including:

    * Post-class thought questions
    * Key terms from the appendix
    * Fundamental concept knowledge

    2. Search for corresponding answers in the textbook

    3. Organize answers into a response database:

    * Direct answers for key terms
    * Guided responses for thought questions
  </Accordion>

  <Accordion icon="code" title="Generate Training Data">
    ```bash theme={null}
    # Navigate to data directory
    cd data

    # Run data generation script
    python generate_data.py
    ```
  </Accordion>
</AccordionGroup>

### Model Fine-tuning

<Steps>
  <Step title="Verify Training Data">
    Check for the presence of training data:

    ```bash theme={null}
    ls -lh data/personal_assistant.json
    ```
  </Step>

  <Step title="Start Fine-tuning">
    Run the training script:

    ```bash theme={null}
    sh train.sh
    ```

    This will use fine-tune the base DeepSeek model.
  </Step>

  <Step title="Check Training Progress">
    Monitor the training directory:

    ```bash theme={null}
    ls -lh train
    ```

    Look for weight directories named `pth_$NUM_EPOCH`
  </Step>

  <Step title="Merge Model Weights">
    ```bash theme={null}
    # Replace $NUM_EPOCH with your target epoch number
    sh merge.sh $NUM_EPOCH
    ```
  </Step>
</Steps>

## Testing and Evaluation

### Interactive Testing

Test your model changes using the chat interface:

```bash theme={null}
sh chat.sh
```

### Automated Evaluation

Run the evaluation suite to measure model performance:

```bash theme={null}
pytest ./test/test_model_evaluation.py
```

This will:

1. Generate responses for test cases
2. Calculate ROUGE similarity scores
3. Output results to `test_results.csv`

## Troubleshooting

<AccordionGroup>
  <Accordion icon="memory" title="GPU Memory Issues">
    If you encounter GPU memory errors:

    1. Reduce batch size in training configuration
    2. Use gradient checkpointing
    3. Ensure no other processes are using GPU memory
  </Accordion>

  <Accordion icon="database" title="Data Generation Issues">
    If data generation fails:

    1. Check input file formats
    2. Verify textbook content is properly formatted
    3. Ensure sufficient disk space
  </Accordion>
</AccordionGroup>

## Contributing

We welcome contributions to InternTA! Please:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

For major changes, please open an issue first to discuss what you would like to change.

## Support

If you need help during development:

1. Check the [GitHub Issues](https://github.com/kongfoo-ai/internTA/issues)
2. Review the [API Documentation](/api-reference/introduction)
3. Contact the development team at [dev@kongfoo.cn](mailto:dev@kongfoo.cn)