bitorch

BITorch is a library currently under development to simplify building quantized and binary neural networks with PyTorch. It contains implementation of the required layers, different quantization functions and examples.

Functions

add_config_args	Adds all arguments from all registered configs.
apply_args_to_configuration	Applies the cli configurations to the config objects.
config_from_name	returns the config to which the name belongs to (name has to be the value of the configs name-attribute)
config_names	Get the list of config names for argparse.
register_custom_config	Register a custom (external) config in bitorch.

Modules

bitorch.config	Config class for bitorch configurations.
bitorch.layers	This submodule contains adapted pytorch layers that use quantization functions on their weights and activations before forwarding them.
bitorch.models	This submodule contains a number of adapted model architectures that use binary / quantized weights and activations.
bitorch.quantizations	This submodule contains several quantization methods that can be used with our quantized layers to build quantized models.
bitorch.runtime_mode
bitorch.util