ok, when we are solve cuda issues,
with torch.no_grad(): seq, alphas, hidden_states = structure_image_beam_search( encoder_out, decoder_structure, word_map_structure, beam_size=args.beam_size_structure)
there is still problem with beam search
ValueError: max() arg is an empty sequence
or is it just too weak model?
write me please if you get something meaningful: kronizzyd@gmail.com
ok, when we are solve cuda issues,
with torch.no_grad(): seq, alphas, hidden_states = structure_image_beam_search( encoder_out, decoder_structure, word_map_structure, beam_size=args.beam_size_structure)
there is still problem with beam searchValueError: max() arg is an empty sequence
or is it just too weak model? write me please if you get something meaningful: kronizzyd@gmail.com