Does Barracuda support quantization model?

[guoguoguilai]

Does Barracuda support quantization model?

[AlexRibard]

Hi!
At the moment, inference is done in floating point precision.
If you import a quanitzed model we will convert it to float.
But you are right, we forgot to handle the UInt8 case. Will be fixed in the upcoming release.

[xunyiljg]

I also encountered this problem and hope to update it as soon as possible. After all, it is more important to greatly optimize the existing model than to support more network architectures.

[mantasp]

@guoguoguilai Uint8 -> float automatic conversion landed to 0.7.0, please let us know if it helped to load your model.