[roedoejet]

PR Goal?

Previously, the loss was calculated by multiplying against the mask, but this would still include the padded frames in MSE loss, unnecessarily. Here we switch to masked select instead to only calculate loss on non-padded inputs/outputs.

Fixes?

Issue raised by personal communication with @korin-richmond

Feedback sought?

sanity

Priority?

medium

Tests added?

yes

How to test?

I've included tests. you should be able to train a new model or finetune existing models without issues

Confidence?

high

Version change?

patch

Related PRs?

EveryVoiceTTS/FastSpeech2_lightning#166

[semanticdiff-com]

Review changes with  

Changed Files

File	Status
everyvoice/model/feature_prediction/FastSpeech2_lightning	0% smaller

[github-actions]

CLI load time: 0:00.24
Pull Request HEAD: 3f62c989ac036ba0466a5618ef8703ab57e8e2b0

Imports that take more than 0.1 s:
import time: self [us] | cumulative | imported package
import time:       698 |     118834 |   typer
import time:      7046 |     241091 | everyvoice.cli
import time:      2121 |     116999 |       markdown_it.main
import time:       408 |     117406 |     markdown_it
import time:      5766 |     168559 |   rich.markdown
import time:      4828 |     285554 | typer.rich_utils

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.25%. Comparing base (a5aedbb) to head (3f62c98).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #887   +/-   ##
=======================================
  Coverage   85.25%   85.25%           
=======================================
  Files          51       51           
  Lines        4789     4789           
  Branches      712      712           
=======================================
  Hits         4083     4083           
  Misses        534      534           
  Partials      172      172           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[joanise]

Oops, you said here you added tests, but they're not there. Did you forget to commit them?

[joanise]

lgtm