search

kimiyoung / transformer-xl

Apache License 2.0
3.61k stars 763 forks source link

What is the meaning of 'bsz' in mem_transformer.py? #116

Closed oshindow closed 4 years ago

oshindow commented 4 years ago

The inputs of Decoderlayer is a set of batch data or a single data ? So as the code in pytorch/mem_trandformer.py, 'qlen=h.size(0)'. But if the input is a batch, h.size(0) is not the qlen.