search

pg020196 / Neural-Network-Translator

This repository contains a neural network translator which can translate neural network models to native code.
MIT License
8 stars 4 forks source link

Plugin possibility at intermediate layer #21

Closed diveflo closed 4 years ago

diveflo commented 5 years ago

It would be awesome if the software architecture would allow for plugins that perform transforms of network at or right before the intermediate layer. Goal: For example, quantization and compression of neural networks, which is especially interesting for resource-constrained target systems. There are framework-specific methods, e.g. tensorflow can output integer-quantized weights, but in the future there might be framework-agnostic algorithms to do this generically. The effort here should hopefully be low as the API should basically just offer “setters” on the intermediate weights and layers and default to an identity operation (==no change).

pg020196 commented 4 years ago

Was implemented in a previous sprint.