Mask for attention #30
Comments
|
I believe this is an issue but is acceptable. Since examples are sorted by the length of the source sentence, during training the length of all source sentences within a batch should be the same for most of the time. Also note that |
marcotcr
pushed a commit
to marcotcr/OpenNMT-py
that referenced
this issue
Sep 20, 2017
Task fixes, changes multiworld.display a tiny bit, added a few init.py's, and fixed a few docs.
marcotcr
pushed a commit
to marcotcr/OpenNMT-py
that referenced
this issue
Sep 20, 2017
* start * batch cleanup, prepare for multiagent batch * oops * generalize display * cleaning single vs. multi + comments * cleaning single vs. multi + comments * cleaning single vs. multi + comments * small update * small fixes * Tasks (OpenNMT#30) Task fixes, changes multiworld.display a tiny bit, added a few init.py's, and fixed a few docs. * multiagent batch * start * batch cleanup, prepare for multiagent batch * oops * generalize display * cleaning single vs. multi + comments * cleaning single vs. multi + comments * cleaning single vs. multi + comments * small update * small fixes * multiagent batch * small * bug
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I wonder why we didn't apply mask in GlobalAttention since we paded zeros on the right. I find there actually have a
applyMaskfunction but not used.Thank you.
The text was updated successfully, but these errors were encountered: