The takeaway: Size is not the Answer

Watch the video outlining the results of our participation in the Biomedical ClinSpEn WMT22 challenge: https://lnkd.in/eN2QNP4K

An excellent experience and milestone achievement!

We took 2nd place among worldwide participants. Initially it was the first place, but later the results were revised.

Lessons learned:
1. Large Language Models do not demonstrate translation quality breakthrough – our winner model was a small model.
2. Cleaning the data carefully before training the model is more important than its size.
3. Automatic “quality” metrics do not show the quality difference between the models.
4. The COMET implementation on the platform is clearly incorrect.

Contact Logrus Global AI Lab for:
1) Winning EN-SP biomedical AI model.
2) Cleaning data to train your custom model.
3) Quality evaluation of the MT output
4) Paralela aligner tool to acquire training data (https://lnkd.in/ejkuf2ve).
5) ..and more.