-
Notifications
You must be signed in to change notification settings - Fork 2
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
There's some ongoing activity trying to repurpose the models for transliteration. Since the models are small, and the authors can eventually end up using them in their day-to-day life since transliteration is viable to have around (in comparison to translation). The following improvements are added: 1. Allow a non-shortlist path 2. Add `nano` config to put in t12n models. This is subject to change during the course of experimentation. 3. Test-suite reading traces from `marian-dev` and generating unit-tests using a Python script (rudimentary, can improve over the course of development). 4. The following bugfixes a. LayerNorm fixed by adjusting default EPS `1e-9` to `1e-6` b. Fix `limit_factor()` narrowed into `size_t` due to mistyping; longer sequences should come now. c. Fix initialization for greedy decode (was not offset correctly before). **Known Issues** 1. `HighwayForward` error at 1e-6 (1e-7 EPS fails), not satisfactory. Abstract operation to be unit-testable. 2. There are still mismatches in the forward--pass to sort out. Pull Request: #47
- Loading branch information
1 parent
3b392f1
commit c20abc9
Showing
32 changed files
with
695 additions
and
991 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.