VishwamAI is an advanced language model based on the Transformer architecture, designed for various natural language processing tasks.
-
Clone the repository:
git clone https://github.com/VishwamAI/chat-agent.git cd chat-agent -
Install the required packages:
pip install -r requirements.txt
To train the VishwamAI model, run:
python scripts/train.py
This script will initialize the model and datasets, and start the training process.
To generate text using a trained model, use:
python scripts/generate_text.py --prompt "Your prompt here" --max_length 100
To evaluate the model on a test dataset, run:
python scripts/evaluate.py --test_file path/to/test/file.txt
To test the VishwamAI model with different sampling parameters (temperature, top-p, top-k) across various prompts, use:
python scripts/sampling_test.py --prompt "Your prompt here" --temperature 0.7 --top_p 0.9 --top_k 50
Before running any scripts that depend on the Hugging Face API, ensure that the HUGGING_FACE_TOKEN environment variable is set up as described in the "Setting Up Environment Variables" section.
You can modify the model and training configuration by editing the configuration files in the configs/ directory.
To securely use the Hugging Face API, set the HUGGING_FACE_TOKEN environment variable with your Hugging Face token. This can be done by adding the following line to your shell profile (e.g., .bashrc, .zshrc):
export HUGGING_FACE_TOKEN=your_hugging_face_token
After adding the line, reload your shell profile:
source ~/.bashrc # or source ~/.zshrc
For more detailed information about the model architecture, training process, and API reference, please refer to the docs/ directory.
Contributions to VishwamAI are welcome! Please refer to the CONTRIBUTING.md file for guidelines on how to contribute to this project.
This project is licensed under the Apache 2.0 License. See the LICENSE file for details.