Ecosyste.ms: OpenCollective

An open API service for software projects hosted on Open Collective.

github.com/mozilla/translations

The code, training pipeline, and models that power Firefox Translations
https://github.com/mozilla/translations

Tests fail in CI

eu9ene opened this issue 7 months ago
Teacher training regression

eu9ene opened this issue 7 months ago
Bump disk for scoring

eu9ene opened this pull request 7 months ago
Error on preemption restart: Not Found model.npz.optimizer.npz

eu9ene opened this issue 7 months ago
"No space left on device" on score step

eu9ene opened this issue 7 months ago
chore: upgrade to Taskgraph 9

bhearsum opened this pull request 7 months ago
Upgrade to Taskgraph 8

bhearsum opened this pull request 7 months ago
Examine strategies for more efficient alignments

gregtatum opened this issue 7 months ago
Close resources in alignments for each step

gregtatum opened this pull request 7 months ago
Add wokers with more memory for translations alignments

gabrielBusta opened this pull request 7 months ago
Update the metrics table published in group_logs

vrigal opened this pull request 7 months ago
Spring 2024 config fixes

gregtatum opened this pull request 7 months ago
Fix data importing for sacrebleu

gregtatum opened this pull request 7 months ago
Fix sacrebleu importing for datasets with the language pair flipped.

gregtatum opened this pull request 7 months ago
Update sv-en-spring-2024.yml

gregtatum opened this pull request 7 months ago
Taskcluster allows max 100 datasets

eu9ene opened this issue 7 months ago
Some mtdata datasets fail because of long name

eu9ene opened this issue 7 months ago
Task has too many dependencies

eu9ene opened this issue 7 months ago
Investigate notifications in Slack

eu9ene opened this issue 7 months ago
Add `aug-mix` to mtdata devtest datasets

gregtatum opened this issue 7 months ago
Fix the monolingual nllb dataset to use zst and report the sentence count

gregtatum opened this pull request 7 months ago
OpusCleaner supports only a limited set of languages

eu9ene opened this issue 7 months ago
Fix creating new projects in W&B

eu9ene opened this pull request 7 months ago
Add mono nllb to the config generation

gregtatum opened this pull request 7 months ago
Add HPLT mono data to the configs

eu9ene opened this pull request 7 months ago
Add HPLT mono bulk importer

eu9ene opened this pull request 7 months ago
Support unexisting project in group_logs publication

vrigal opened this pull request 7 months ago
Update the config generator to remove 404s, and improve the Content-Length lookup

gregtatum opened this pull request 7 months ago
Exclude mtdata datasets from config generation that return a 404

gregtatum opened this issue 7 months ago
Fix config generation for ted talks and hide known inaccurate sizes

gregtatum opened this pull request 7 months ago
clean corpus unable to fork sometimes on generic-worker/d2g

bhearsum opened this issue 7 months ago
Create a new W&B project if it doesn't exist

eu9ene opened this issue 7 months ago
Fix offline group publication

vrigal opened this pull request 7 months ago
Parse stalled validation data

vrigal opened this pull request 7 months ago
Prepare configs for training

eu9ene opened this pull request 8 months ago
Figure out accuracy of mtdata_Neulab-tedtalks datasets

gregtatum opened this issue 8 months ago
Figure out sacrebleu `/` dataset strategy

gregtatum opened this issue 8 months ago
[meta] Make config generation fully automated

gregtatum opened this issue 8 months ago
In config generation, switch to two stage training when the mono data is too small

gregtatum opened this issue 8 months ago
bicleaner-ai-classify intermittently fails to download fasttext model

eu9ene opened this issue 8 months ago
Many random errors on CPU generic workers

eu9ene opened this issue 8 months ago
Revert CPU generic workers

eu9ene opened this pull request 8 months ago
Evaluation fails on a pre-trained backward model

eu9ene opened this issue 8 months ago
Avoid publishing experiment config when W&B publication is disabled

vrigal opened this pull request 8 months ago
Avoid publishing experiment config when W&B publication is disabled

vrigal opened this pull request 8 months ago
actions don't work in task groups created from another action

bhearsum opened this issue 8 months ago
Retry tasks on exit code 128 to work around intermittent issues with DNS failures

bhearsum opened this pull request 8 months ago
previous_group_ids needs to gracefully handle upstream tasks being expired

bhearsum opened this issue 8 months ago
Update Flores-101 data importer to the Flores-200 dataset

gregtatum opened this issue 8 months ago
Publish comet metrics

vrigal opened this pull request 8 months ago
Automatically generate training config files with the `task config-generator`

gregtatum opened this pull request 8 months ago
WIP: try to use d2g for train-backwards

bhearsum opened this pull request 8 months ago
bring snakepit machines online with taskcluster

bhearsum opened this issue 8 months ago
Switch OpusTrainer log level to info

eu9ene opened this issue 8 months ago
Support `.metrics` files in GCP experiments publication

vrigal opened this issue 8 months ago
Support COMET metric in the parser

vrigal opened this issue 8 months ago
task lint-fix is still broken for me locally for formatting imports

gregtatum opened this issue 8 months ago
Build and use cyhunspell binary wheel

bhearsum opened this pull request 8 months ago
Ensure all of the task labels can be parsed in the task graph

gregtatum opened this pull request 8 months ago
Use shorter names for dataset URLs

gregtatum opened this pull request 8 months ago
Use a defined run ID on W&B (refactoring)

vrigal opened this issue 8 months ago
Parse tasks with label finetune-student

vrigal opened this pull request 8 months ago
Support news-crawl importer

eu9ene opened this pull request 8 months ago
Consolidate all of the training scripts into a main pipeline/train/train.py script

gregtatum opened this issue 8 months ago
Change the caching strategy for teacher ensembles

gregtatum opened this pull request 8 months ago
finetune-student died when publication failed

bhearsum opened this issue 8 months ago
Upgrade to Bicleaner 3

eu9ene opened this pull request 8 months ago
Add news-crawl importer to find-corpus

eu9ene opened this issue 8 months ago
Publish experiment config from taskcluster training task (group_logs)

vrigal opened this pull request 8 months ago
Address issues with resuming training in W&B

eu9ene opened this issue 8 months ago
Our models should be robust enough to translate a calendar

gregtatum opened this issue 8 months ago
Rewrite the training script in Python

eu9ene opened this issue 8 months ago
Publish evaluation metrics

vrigal opened this pull request 8 months ago
Consolidate yaml schema and configs

eu9ene opened this issue 8 months ago
Add ability to switch to a one-stage teacher training

eu9ene opened this pull request 8 months ago
Override W&B data on a resumed training

vrigal opened this pull request 8 months ago
weights and biases reporting doesn't work correctly when resuming training after a spot termination

bhearsum opened this issue 8 months ago
Publish evaluation metrics (rebase)

vrigal opened this pull request 8 months ago
Investigate the effect of back-translations

eu9ene opened this issue 8 months ago
Migrate all the run_task tests into a separate folder

gregtatum opened this issue 8 months ago
[wip] add support for automatic uploading of artifacts (fixes #466)

bhearsum opened this pull request 8 months ago
Generic Taskcluster task naming

vrigal opened this pull request 8 months ago
Requirements files should be a key in the kind.yml, and they should be installed with a transform

gregtatum opened this issue 8 months ago
Add COMET to the evaluation

gregtatum opened this pull request 8 months ago
TaskCluster docs for non-maintainers

AmitMY opened this issue 8 months ago
Fix W&B publication setting

eu9ene opened this pull request 8 months ago
Config setting wandb-publication doesn't work

eu9ene opened this issue 8 months ago
Fix config parsing

eu9ene opened this pull request 8 months ago
[taskcluster:error] Error uploading "public/build/tmp/aln.fwd" artifact. ext.certificate.expiry < now

eu9ene opened this issue 8 months ago
Explore using genetic algorithms for data cleaning rules

eu9ene opened this issue 8 months ago
Add support for automatically continuing training from earlier runs of a Task (fixes #270)

bhearsum opened this pull request 8 months ago
Get rid of train-taskcluster.sh

bhearsum opened this issue 8 months ago
Add an author field

eu9ene opened this issue 8 months ago
Ensure numbers associated to units of measure are not translated as the same number with another unit of measure

marco-c opened this issue 8 months ago
Fix mono downloader

eu9ene opened this pull request 8 months ago
group_logs is incomplete

eu9ene opened this issue 8 months ago
Actualize moz-translations when online uploading is feature complete

eu9ene opened this issue 8 months ago
Ensure naming for runs is consistent

eu9ene opened this issue 8 months ago
Add training configs for prod models

eu9ene opened this issue 8 months ago
Report empty alignments separately

eu9ene opened this pull request 8 months ago