Closed leonhardhennig closed 2 years ago
It seems like the new tokenizer from huggingface are not compatible with python 3.9.7 - but work with python 3.8. Tested on 2 machines. Would be great if somebody can confirm. Enroot images do not have issues, as they are all using python 3.8 (if I am not mistaken).
Installing rust solves the issue: curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
link1, link2
After minor adaptations the repository works good in python 3.8 with pytorch 1.10 (latest)
Relation classification:
f1 = 0.7062233068944478
precision = 0.7164964407304241
recall = 0.6962406015037594
Named Entitiy Recognition
f1 = 0.8298232372419166
loss = 0.21359181425055318
precision = 0.8428022107098257
recall = 0.8172379479432675
and test, e.g. if issue #29 works
when done, merge into branch sherlock_version_2