multimodel-ai 0.1.2

A Python module for efficient multi-model AI inference with memory management

MultiModel-AI

A Python module for efficient multi-model AI inference with memory management.

Features

• Efficient memory management for multiple AI models
• Automatic model loading and unloading
• Support for various AI model types
• Easy-to-use API

Installation

pip install multimodel-ai

Usage

from multimodel_ai import ModelManager

# Initialize the model manager
manager = ModelManager()

# Load a model
model = manager.load_model("model_name")

# Use the model
result = model.predict(input_data)

# The model will be automatically unloaded when not in use

Examples

Basic Usage

from multimodel_ai import ModelManager

# Initialize the manager
manager = ModelManager()

# Load a model
model = manager.load_model("gpt2")

# Generate text
text = model.generate("Hello, world!")

# The model will be automatically unloaded

Multiple Models

from multimodel_ai import ModelManager

# Initialize the manager
manager = ModelManager()

# Load multiple models
model1 = manager.load_model("gpt2")
model2 = manager.load_model("bert")

# Use the models
text1 = model1.generate("Hello")
text2 = model2.classify("World")

# Models will be automatically unloaded when not in use

Custom Model Configuration

from multimodel_ai import ModelManager

# Initialize the manager with custom settings
manager = ModelManager(
    max_memory_usage=0.8,  # Use up to 80% of available memory
    model_cache_dir="./models"  # Custom cache directory
)

# Load a model with specific configuration
model = manager.load_model(
    "gpt2",
    device="cuda",
    precision="fp16"
)

# Use the model
result = model.generate("Hello, world!")

Error Handling

from multimodel_ai import ModelManager, ModelError

try:
    manager = ModelManager()
    model = manager.load_model("non_existent_model")
except ModelError as e:
    print(f"Error loading model: {e}")

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.