flextensor 0.1.0

FlexTensor: Tensor offloading and management library

FlexTensor

FlexTensor is a tensor offloading and management library for PyTorch that enables running large models on limited GPU memory by intelligently offloading tensors between GPU and CPU memory.

Features

• Simplified API: Easy-to-use high-level API for automatic tensor offloading
• Automatic Model Patching: Offload model layers without modifying model code
• Manual Control: Fine-grained control with offload_block context managers
• Smart Profiling: Automatic warmup and profiling for optimal performance
• Wildcard Support: Use patterns like "layers.*" to offload multiple modules
• Profile Persistence: Save and load offloading profiles for faster startup
• Lazy Model Initialization: Load models from saved profiles with optimized weight loading
• Shared Memory: Optional shared memory subsystem for cross-process tensor coordination

Documentation

For detailed guides, API reference, and more, visit our Documentation.

Quick Installation

To install FlexTensor from PyPI:

pip install flextensor

For more installation options (source, dev, optional dependencies), see the Installation Guide.

Quick Example

import flextensor
from flextensor import OffloadConfig

# Your existing model
model = YourModel()

# Configure offloading
config = OffloadConfig(
    gpu_device=0,              # GPU to use
    warmup_iters=1,            # Iterations for tensor discovery
    profile_iters=10,          # Iterations for timing measurement
    module_patterns=["layers.*"],  # Which modules to offload
)

# Patch the model
model = flextensor.offload(model, config=config)

# Use normally - first warmup_iters + profile_iters iterations are warmup/profile
for batch in dataloader:
    output = model(batch)  # FlexTensor handles everything

See the Quick Start for more examples.

License

FlexTensor is licensed under the Apache License 2.0.