capibara-model 1.1.5

Language model with contextual processing

🦫 CapibaraModel

Language model with contextual processing based on JAX/Flax.

✨ Features

• 🧠 Architecture:

  • 🔄 Multi-head attention
  • 🎯 Contextual activation
  • 🔍 Coherence detection
  • 🎭 Personality management

• ⚡ Optimizations:

  • 🚀 Native TPU support
  • ⚙️ Efficient processing
  • 📦 Optimized batching
  • 🕸️ Integrated sparsity

📋 Requirements

• Python >= 3.8
• JAX >= 0.4.1
• Flax >= 0.6.0
• Optax >= 0.1.3

🚀 Installation

pip install capibara_model

💻 Usage

Basic Example

from capibara_model.core.model import CapibaraModel
from capibara_model.core.inference import CapibaraInference

# Create model
model = CapibaraModel(hidden_size=768, num_heads=8)

# Create inference
inference = CapibaraInference(hidden_size=768)

# Generate response
response = inference("How are you?")
print(response)

Advanced Example

# Custom configuration
config = {
    'model': {
        'hidden_size': 1024,
        'num_heads': 16,
        'num_layers': 24,
        'dropout_rate': 0.1
    },
    'training': {
        'batch_size': 32,
        'learning_rate': 1e-4,
        'warmup_steps': 1000
    }
}

# Create model with custom config
model = CapibaraModel(**config['model'])

# Process conversation
context = "Previous conversation context..."
response = inference(
    "What's the meaning of life?",
    context=context
)

⚙️ Configuration

config = {
    'hidden_size': 768,
    'num_heads': 8,
    'num_layers': 12,
    'dropout_rate': 0.1,
    'use_tpu': False
}

🛠️ Development

# Install development dependencies
pip install capibara_model[dev]

# Run tests
pytest tests/

# Run specific test
pytest tests/test_model.py -k "test_attention"

# Check code style
black capibara_model/

📝 License

MIT License. See LICENSE for more information.

📫 Contact

Marco Durán - [e-mail]

🌐 Website | 💻 GitHub

🙏 Acknowledgments

Special thanks to:

• JAX/Flax team
• TPU Research Cloud
• Open source community