Today
-
Package G: 6 BERTs (5 MLM-pretrained + 1 FFN-widened MLM-pretrained), knn 0.1, mlp 0.3
2 days ago
-
Package E: 7 MLM-pretrained BERTs (4 with pseudo-labels), knn 0.1, mlp 0.3
3 days ago
-
Package D: 5 MLM-pretrained BERTs (2 with pseudo-labels round 2) + pseudo model, knn 0.1, mlp 0.3
6 days ago
-
Package C: MLM-pretrained BERTs (3) + full120/80 models, knn 0.1, mlp 0.3
8 days ago
-
Final: 120-epoch models w1,w2 + p1 + r1 + l1 + l2, kNN 0.1, MLP 0.3
9 days ago
-
blend variant v3_knn0.25_mlp0.4
-
blend variant v2_knn0.1_mlp0.45
-
blend variant v1_knn0.2_mlp0.3
-
Package B: round-9 pseudo-label 80-epoch models + p1 + l1 + l2 + r1, kNN 0.1, MLP 0.3
10 days ago
-
Package A: pseudo model + 80-epoch models + all-data BERTs, kNN 0.1, MLP 0.3
-
6 BERTs incl. pseudo-label model (replaces full_s1) + kNN 0.1 + MLP 0.3
11 days ago
-
6 BERTs incl. random-SMILES-augmented model + kNN 0.1 + MLP 0.3
-
6 all-data BERTs + kNN 0.1 + MLP 0.3
-
Final ensemble: 4 all-data BERTs + kNN 0.1 + MLP 0.3
12 days ago
-
Interim blend kNN 0.3
-
PROBE BERT only
-
PROBE kNN only
-
Interim: 3 BERTs (lr1e-4 40ep, 25ep, B) int8 + kNN w=0.3
-
Ensemble: 2 full-data BERTs (25 ep, aug) int8 + TF-IDF kNN
14 days ago
-
Ensemble of 2 reaction BERTs (int8) + TF-IDF kNN
19 days ago
-
Reaction BERT classifier (8 epochs, full train data)
-
initial commit
