Closed xu-gaopeng closed 1 year ago
@xu-gaopeng Thanks for the catch, these are obviously typos.
~What's odd, the config seems not to have an output-xent
layer. This is off. Have you ran the trainer, at least for a few iterations, just to test it?~ I was thinking this used the chain objective, sorry. Please ignore.
input=Append(blstm3-forward, blstm3-backward)
change makes sense, though,
or not?
y.
On Wed, Sep 14, 2022 at 11:06 AM Daniel Povey @.***> wrote:
@.**** commented on this pull request.
In egs/aspire/s5/local/nnet3/run_blstm.sh https://github.com/kaldi-asr/kaldi/pull/4790#discussion_r970531121:
- output-layer name=output output-delay=$label_delay dim=$num_targets max-change=1.5
- output-layer name=output input=Append(blstm3-forward, blstm3-backward) output-delay=$label_delay dim=$num_targets max-change=1.5
0 is the default so it's not really important here.
— Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/pull/4790#discussion_r970531121, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACUKYX5BJUT2JI2RHUXMCRDV6GIQNANCNFSM6AAAAAAQGXD52A . You are receiving this because you were assigned.Message ID: @.***>
@jtrmal Yes, this one certainly does. The default is the previous line's output.
Thanks, merging. Sorry for the delay! y.
fix bug aspire run_blstm.sh