Edit model card

norbert3-small trained on wikiann (fo/is), sucx3 (se), dane (da) and norne (nb/nn)

added a custom clf head along with a character-level cnn for adding a tiny extra signal for the classification.

results:

Eval on wikiann - fo
index                                  0
tokens      [Byrta, -, Aftur, og, aftur]
ner_tags                 [3, 0, 0, 0, 0]
subset                                fo
dataset                          wikiann
Name: 0, dtype: object
shape: (100, 5)
100%5/5 [00:01<00:00,  3.92it/s]
Loss: 0.2276667356491089
O	O
B-ORG	B-ORG
B-ORG	B-ORG
O	O
O	O
O	O
O	O
O	O
O	O
O	O
Validation Loss: 0.26530784368515015
Validation Accuracy: 0.9228951181745751
              precision    recall  f1-score   support

         LOC       0.86      0.81      0.83       154
         ORG       0.67      0.73      0.70       125
         PER       0.87      0.91      0.89        79

   micro avg       0.79      0.80      0.80       358
   macro avg       0.80      0.82      0.81       358
weighted avg       0.79      0.80      0.80       358

________________________________________
Eval on wikiann - is
index                                                 100
tokens      [Beltaþyrill, ''Ceryle, alcyon, '', Sjaldséð]
ner_tags                                  [5, 0, 0, 0, 0]
subset                                                 is
dataset                                           wikiann
Name: 0, dtype: object
shape: (1000, 5)
100%50/50 [00:10<00:00,  5.02it/s]
Loss: 0.22668001055717468
O	O
B-LOC	B-LOC
B-LOC	B-LOC
B-LOC	B-LOC
B-LOC	B-LOC
B-LOC	B-LOC
B-LOC	B-LOC
O	O
O	O
O	O
Validation Loss: 0.2526825902983546
Validation Accuracy: 0.9360383541181041
              precision    recall  f1-score   support

         LOC       0.84      0.85      0.84      1983
         ORG       0.81      0.80      0.80      1762
         PER       0.89      0.89      0.89      1020

   micro avg       0.84      0.84      0.84      4765
   macro avg       0.84      0.85      0.85      4765
weighted avg       0.84      0.84      0.84      4765

________________________________________
Eval on dane - default
index                                                    1100
tokens      [To, kendte, russiske, historikere, Andronik, ...
ner_tags    [0, 0, 7, 0, 1, 2, 0, 1, 2, 0, 0, 0, 0, 5, 0, ...
subset                                                default
dataset                                                  dane
Name: 0, dtype: object
shape: (565, 5)
100%29/29 [00:06<00:00,  4.75it/s]
Loss: 0.12037135660648346
O	O
O	O
O	O
O	O
O	O
B-MISC	B-MISC
O	O
O	O
B-PER	B-PER
B-PER	B-PER
Validation Loss: 0.11113663488228259
Validation Accuracy: 0.972018408457994
              precision    recall  f1-score   support

         LOC       0.78      0.86      0.82       225
        MISC       0.72      0.52      0.61       333
         ORG       0.72      0.69      0.71       379
         PER       0.96      0.92      0.94       298

   micro avg       0.80      0.73      0.76      1235
   macro avg       0.80      0.75      0.77      1235
weighted avg       0.79      0.73      0.76      1235

________________________________________
Eval on norne - bokmaal-7
index                                                    1665
tokens      [Honnørordene, er, ", dristig, formspråk, ", ,...
ner_tags     [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
subset                                              bokmaal-7
dataset                                                 norne
Name: 0, dtype: object
shape: (1939, 5)
100%97/97 [00:20<00:00,  4.56it/s]
Loss: 0.0011819382198154926
O	O
O	O
O	O
O	O
O	O
O	O
O	O
O	O
O	O
O	O
Validation Loss: 0.04194018930858649
Validation Accuracy: 0.9876322465792248
              precision    recall  f1-score   support

         LOC       0.85      0.90      0.87       498
        MISC       0.81      0.74      0.78       363
         ORG       0.77      0.83      0.80       499
         PER       0.93      0.96      0.95       845

   micro avg       0.86      0.88      0.87      2205
   macro avg       0.84      0.86      0.85      2205
weighted avg       0.86      0.88      0.87      2205

________________________________________
Eval on norne - nynorsk-7
index                                                    3604
tokens      [Den, er, mettande, og, smakfull, ,, og, det, ...
ner_tags    [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, ...
subset                                              nynorsk-7
dataset                                                 norne
Name: 0, dtype: object
shape: (1511, 5)
100%76/76 [00:15<00:00,  5.82it/s]
Loss: 0.0790824368596077
O	O
O	O
O	O
O	O
O	O
O	O
O	O
O	O
O	O
O	O
Validation Loss: 0.05325472676725583
Validation Accuracy: 0.9867293689853402
              precision    recall  f1-score   support

         LOC       0.77      0.91      0.84       365
        MISC       0.80      0.76      0.78       295
         ORG       0.83      0.82      0.82       397
         PER       0.98      0.95      0.97       664

   micro avg       0.87      0.88      0.87      1721
   macro avg       0.85      0.86      0.85      1721
weighted avg       0.87      0.88      0.87      1721

________________________________________
Eval on sucx3_ner - original_cased
index                                                    5115
tokens      [Just, i, dag, är, Saabs, företagsledning, där...
ner_tags    [0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, ...
subset                                         original_cased
dataset                                             sucx3_ner
Name: 0, dtype: object
shape: (14383, 5)
100%720/720 [02:36<00:00,  5.02it/s]
Loss: 0.04177908971905708
Loss: 0.08230985613484489
Loss: 0.08399457804886486
Loss: 0.06163447560524267
Loss: 0.04787629511204947
Loss: 0.03949779063830233
Loss: 0.03397762095776484
Loss: 0.030040143460689266
O	O
O	O
O	O
O	O
O	O
B-ORG	B-ORG
B-ORG	B-ORG
O	O
O	O
O	O
Validation Loss: 0.02938824465528948
Validation Accuracy: 0.9919830972756728
              precision    recall  f1-score   support

         LOC       0.88      0.91      0.90      4202
        MISC       0.65      0.59      0.62      1899
         ORG       0.74      0.76      0.75      3015
         PER       0.92      0.93      0.92      5778

   micro avg       0.84      0.84      0.84     14894
   macro avg       0.80      0.80      0.80     14894
weighted avg       0.84      0.84      0.84     14894

________________________________________
Downloads last month
5
Safetensors
Model size
45.4M params
Tensor type
I64
·
F32
·
Inference API
Unable to determine this model’s pipeline type. Check the docs .