Greg Tatum

Requirements files should be a key in the kind.yml, and they should be installed with a transform

1

This will help with our local testing config, and simplify our kind.yml files. Something like: ``` diff --git a/taskcluster/kinds/evaluate-teacher-ensemble/kind.yml b/taskcluster/kinds/evaluate-teacher-ensemble/kind.yml index cb945c5..0bace02 100644 --- a/taskcluster/kinds/evaluate-teacher-ensemble/kind.yml +++ b/taskcluster/kinds/evaluate-teacher-ensemble/kind.yml @@ -22,6 +22,7...

The training_continuation.py doesn't correctly handle multiple URLs. Steps to reproduce: Set the config ``` experiment: pretrained-models: train-teacher: urls: - https://example.com/a - https://example.com/b mode: use type: default ``` Run `task preflight-check`...

We already have OpusCleaner defaulting to true on the production config, but after some time we should go ahead and delete the old cleaning procedures.

Add file overrides to the training continuation, and refactor the implementation

2

Edit: I added file overrides for this, to support vocabs and other file name mismatches. For instance in `en-fi` the final model was not a `best-chrf` but a `best-perplexity`. This...

The data was produced from web crawls, and has a cleaned version of the data. It includes language detection via [FastSpell](https://github.com/mbanon/fastspell) (a combo of FastText and Hunspell). It also includes...

RTL languages shouldn't affect training, but doing so will require some work on the Firefox side. This meta bug tracks any work that is needed. We should complete a subset...

[meta] Train easy to segment LTR languages

2

In the short term we are focusing on building up our language list by training easy to segment LTR languages, as they don't require changes to the training pipeline, and...

Update home screen UI for loading existing profiles from pasting

22

Right now we only allow for loading in a profile by file:  It would be nice to have a button that say something like "Load existing profile" that when...

In PR #3051, I removed most of the existential types from our codebase, but there are some in our higher order components (HOC). It would be nice to remove all...

Right now we have lots of custom handling. This could be done purely with marker schema. e.g. for Network markers: ``` if (newData.type === 'Network') { if (typeof newData.domainLookupStart ===...