Oktai Tatanov

@tuanshanyou , you can try this new library: https://github.com/VKCOM/YouTokenToMe. This library is more efficient than fastBPE.

@learning2hash generator has around 2.2 millions

@cndn I see that pytorch-translate has version of BeamSearchV2 that have new improvements, but it doesn't support char source. Why?

@m-toman yes, NeMo code is updated to TalkNet2

@Xiaodong-Bran, did you understand this way?

@Jongchan, hi! Will pretrained model be released?

Hello! Will you plan to add Objective/Subjective evaluation?

@CookiePPP great, thanks! Then I need number of total batch size that was used :)

@mdbloice but if I generate ground truth on the fly? How can I pass it in Pipeline so that image and mask will be transformed equally?

@pyeguy @thegodone in jtnn_vae.py replace: `if len(labels) == 0: return create_var(torch.Tensor(0)), 1.0` with `if len(labels) == 0: return create_var(torch.Tensor([0])), 1.0`