4 Pull requests merged by 4 people

• feat: upgrade cpu workers to ubuntu 24.04 generic-worker image

  #949 merged Dec 17, 2024

• Rollback target renaming

  #960 merged Dec 16, 2024

• Fix Space Omission for Korean Translations

  #958 merged Dec 16, 2024

• Add a warning when train- is the target stage rather than evaluate-

  #951 merged Dec 13, 2024

5 Pull requests opened by 3 people

• [Experiment] CJK with ICU segmenter

  #959 opened Dec 16, 2024

• Rename "all" to "all-pipeline"

  #962 opened Dec 17, 2024

• Do not output empty alignments

  #963 opened Dec 17, 2024

• feat: add cron task that runs the minimal training pipeline nightly

  #966 opened Dec 18, 2024

• Remove expired-after

  #968 opened Dec 18, 2024

3 Issues closed by 2 people

• quantize often dumps core since browsermt-marian was updated

  #281 closed Dec 17, 2024

• evaluate-quantized step fails in CI

  #454 closed Dec 17, 2024

• Autogenerated config doesn't work

  #941 closed Dec 16, 2024

5 Issues opened by 2 people

• Update dependencies to upstream OpusTrainer when released

  #967 opened Dec 18, 2024

• Investigate translating mixed Simplified/Traditional script

  #965 opened Dec 18, 2024

• Error on resuming training after preemption

  #964 opened Dec 17, 2024

• Errors in Marian logs parser

  #961 opened Dec 17, 2024

• Investigate marian-decoder memory usage

  #957 opened Dec 13, 2024

12 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Rewrite the train scripts and add config support for ctranslate2

  #922 commented on Dec 18, 2024 • 7 new comments

• Add a util for local remote settings

  #955 commented on Dec 12, 2024 • 2 new comments

• Switch to ICU tokenizer

  #939 commented on Dec 19, 2024 • 1 new comment

• scrape and upload 2024 training artifacts

  #948 commented on Dec 13, 2024 • 0 new comments

• Investigate switching to ICU segmenter

  #860 commented on Dec 16, 2024 • 0 new comments

• Japanese is missing in OpusCleaner

  #943 commented on Dec 17, 2024 • 0 new comments

• CI often fails with "Could not resolve host: github.com"

  #549 commented on Dec 17, 2024 • 0 new comments

• Single-side deduplication

  #928 commented on Dec 17, 2024 • 0 new comments

• [meta] Kick off a 2024-H2 training run

  #912 commented on Dec 17, 2024 • 0 new comments

• previous_group_ids needs to gracefully handle upstream tasks being expired

  #623 commented on Dec 17, 2024 • 0 new comments

• [meta] Train harder to segment languages, like CJK languages

  #425 commented on Dec 18, 2024 • 0 new comments

• WIP switch GPU workers to image that uses multi engine generic worker

  #700 commented on Dec 19, 2024 • 0 new comments