Open lee-hodg opened 2 years ago
Hello,
Currently all our models are trained with full precision so reducing the precision only at inference time could be unsafe. However, it should be possible and the torch docs on quantization should be helpful. We are planning to look more into a mobile version of Detoxify in the future, and of course, contributions are welcome!
Is there anyway I can load the weights and "quantize" to reduce the size so I can run with pytorch mobile client side?