Update Model
#3
by stefan-it - opened
Hi,
this PR adds a retrained version of the upos-multi model.
The retrained version fixes this bug reported in Flair upstream repo :)
stefan-it changed pull request status to open
Pinging @alanakbik :)
Notice: the retrained model does not introduce a performance degradation: in the old readme a wrong accuracy score was displayed - this can be seen in the old training.log file. Old accuracy was 96.41, new accuracy is 96.87!
alanakbik changed pull request status to merged
@stefan-it I see different results for the old vs new model for french. Is there a way to update the weights without retraining to make them compatible?