search

vvolhejn / thesis

ETH Zürich MSc Thesis: Accelerating Neural Audio Synthesis
Apache License 2.0
17 stars 1 forks source link

Choose a framework - Pytorch vs TensorFlow #2

Closed vvolhejn closed 2 years ago

vvolhejn commented 2 years ago

framework-wise, I would go for Tensorflow because a) optimization (quantization+pruning) seems to be mature https://www.tensorflow.org/lite/performance/post_training_integer_quant whereas Pytorch still calls quantization beta https://pytorch.org/tutorials/advanced/static_quantization_tutorial.html b) the DDSP library https://github.com/magenta/ddsp which the original model is based on uses TF as well

Also there is TFLite.

vvolhejn commented 2 years ago

Tensorflow seems to be the way to go, but it is a bit sad because most other implementations apart from DDSP use PyTorch.