mna-acquisition-binary / tokenizer.json

Commit History

v6-maxlen256: L-2_H-512_A-8 retrained on corrected dataset (label fixes + non-ASCII text cleanups on 3 long articles). Test F1=0.9424, Recall=0.9796, Precision=0.7869, F2=0.9339, threshold=0.16, CPU 30.22ms.
dcbfb3d
verified

alexge233 commited on

v5 binary acquisition classifier — google/bert_uncased_L-2_H-512_A-8, MAX_LENGTH=256, Optuna trial 61, test F1 macro 0.8813, recall(1) 0.9796 @ threshold 0.26, ~24ms CPU inference
e098f47
verified

alexge233 commited on