gordicaleksa/Open-NLLB

Issues

• Standard Moroccan Tamazight is mislabeled

  #28 opened 8 months ago by MedAymenF
  0

• Provide context for input and output

  #29 opened 8 months ago by joiemoie
  0

• Creation of a small model file for a few languages

  #27 opened 10 months ago by dlippold
  0

• MinHash: benchmark memory, speed and accuracy with varying r and b

  #23 opened a year ago by vienneraphael
  1

• Native language visualizations

  #6 opened a year ago by vienneraphael
  2

• Weird line length spikes in Serbian, Croatian, Bosnian (data analysis task)

  #22 opened a year ago by gordicaleksa
  4

• [Modeling] Release a 615M HBS (Croatian, Bosnian, Serbian) Open-NLLB checkpoint

  #15 opened a year ago by gordicaleksa
  0

• [Data] Acquire additional high-quality (non-public) parallel corpora for HBS

  #19 opened a year ago by gordicaleksa
  0

• [Future - outside current project scope] non-English LLMs (Serbian LLM, etc.)

  #21 opened a year ago by gordicaleksa
  0

• [Future - outside current project scope] 7B lang-family-specific Open-NLLB checkpoint

  #20 opened a year ago by gordicaleksa
  0

• [Modeling] Release a 1.3B Slavic languages Open-NLLB checkpoint

  #16 opened a year ago by gordicaleksa
  0

• [Modeling] Release a 3.3B Open-NLLB checkpoint (~202 languages)

  #17 opened a year ago by gordicaleksa
  0

• [Modeling] Release a 615M English -> HBS Open-NLLB checkpoint

  #18 opened a year ago by gordicaleksa
  0

• Get a compute grant

  #14 opened a year ago by gordicaleksa
  0

• Understand how to do 4-stage curriculum learning from the paper

  #12 opened a year ago by gordicaleksa
  0

• sub-batches creation

  #4 opened a year ago by vienneraphael
  1

• Estimate the necessary compute and number of GPUs for Open-NLLB effort

  #13 opened a year ago by gordicaleksa
  0

• Reduce peak memory when using FSDP on 2+ GPUs

  #2 opened a year ago by gordicaleksa
  0

• Obtain high quality Serbian parallel corpus (currently 0 support in our public bi-text)

  #10 opened a year ago by gordicaleksa
  0

• Setup a pipeline for mined data (use Allen AI's OSS dataset replication)

  #11 opened a year ago by gordicaleksa
  0

• Hydra pickle issue in generate_multi.py

  #3 opened a year ago by vienneraphael
  0

• Spanish and Guarani filtering

  #5 opened a year ago by vienneraphael
  1

• Choosing the LID model

  #8 opened a year ago by vienneraphael
  0

• LID model peak probabilities

  #7 opened a year ago by vienneraphael
  0