Models proposed to test - Githubissues

CLARIN-PL / LEPISZCZE

This is the way: designing and compiling LEPISZCZE, a comprehensive NLP benchmark for Polish

MIT License

13 stars 2 forks source link

Models proposed to test #5

Open djstrong opened 2 years ago

djstrong commented 2 years ago

It would be helpful to see those models from HF on the leaderboard:

xlm-roberta-base (base of HerBERT)
xlm-roberta-large (base of HerBERT)
facebook/xlm-roberta-xl - needs more VRAM
facebook/xlm-roberta-xxl - needs more VRAM
sdadas/polish-distilroberta - something smaller
sdadas/polish-roberta-base-v2 - maybe better than HerBERT
sdadas/polish-roberta-large-v2 - maybe better than HerBERT

Other models:

Flair
word2vec
GloVe
FastText
some non-vector based models from scikit-learn, e.g NB, SVM, XGB, logistic regression

laugustyniak commented 2 years ago

Hi @djstrong, thanks for your recommendation.

Right now, we are gathering baseline models for testing, we already added to our todo list static embeddings with SVM/XGB.

Regarding LMs, we also plan to add them but rather in 3rd batch of our experiment running.