Hugging Face's logo

stulcrad
/
CNEC_2_0_Supertypes_slavicbert

Token Classification
Transformers
Safetensors
bert
Generated from Trainer
Eval Results
Inference Endpoints
Model card Files Community
stulcrad commited on
Commit
d38c73a
·
verified ·
1 Parent(s): 7880a4d

Model save

Browse files
Files changed (1) hide show
  1. README.md +19 -28
README.md CHANGED
@@ -24,16 +24,16 @@ model-index:
24
  metrics:
25
  - name: Precision
26
  type: precision
27
- value: 0.8392434988179669
28
  - name: Recall
29
  type: recall
30
- value: 0.8798017348203222
31
  - name: F1
32
  type: f1
33
- value: 0.8590441621294617
34
  - name: Accuracy
35
  type: accuracy
36
- value: 0.9608422328258729
37
  ---
38
 
39
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
@@ -43,11 +43,11 @@ should probably proofread and complete it, then remove this comment. -->
43
 
44
  This model is a fine-tuned version of [DeepPavlov/bert-base-bg-cs-pl-ru-cased](https://huggingface.co/DeepPavlov/bert-base-bg-cs-pl-ru-cased) on the cnec dataset.
45
  It achieves the following results on the evaluation set:
46
- - Loss: 0.1943
47
- - Precision: 0.8392
48
- - Recall: 0.8798
49
- - F1: 0.8590
50
- - Accuracy: 0.9608
51
 
52
  ## Model description
53
 
@@ -72,30 +72,21 @@ The following hyperparameters were used during training:
72
  - seed: 42
73
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
74
  - lr_scheduler_type: linear
75
- - num_epochs: 20
76
 
77
  ### Training results
78
 
79
  | Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
80
  |:-------------:|:-----:|:----:|:---------------:|:---------:|:------:|:------:|:--------:|
81
- | 0.5752 | 1.11 | 500 | 0.2395 | 0.6751 | 0.7381 | 0.7052 | 0.9324 |
82
- | 0.2671 | 2.22 | 1000 | 0.1882 | 0.7554 | 0.8253 | 0.7888 | 0.9476 |
83
- | 0.2057 | 3.33 | 1500 | 0.1645 | 0.7985 | 0.8563 | 0.8264 | 0.9533 |
84
- | 0.1713 | 4.44 | 2000 | 0.1852 | 0.7936 | 0.8542 | 0.8228 | 0.9531 |
85
- | 0.1562 | 5.56 | 2500 | 0.1724 | 0.8051 | 0.8583 | 0.8309 | 0.9550 |
86
- | 0.1289 | 6.67 | 3000 | 0.1639 | 0.8203 | 0.8711 | 0.8450 | 0.9598 |
87
- | 0.1186 | 7.78 | 3500 | 0.1763 | 0.8216 | 0.8691 | 0.8446 | 0.9585 |
88
- | 0.1031 | 8.89 | 4000 | 0.1764 | 0.8267 | 0.8707 | 0.8481 | 0.9584 |
89
- | 0.0973 | 10.0 | 4500 | 0.1868 | 0.8330 | 0.8798 | 0.8558 | 0.9597 |
90
- | 0.0877 | 11.11 | 5000 | 0.1818 | 0.8304 | 0.8835 | 0.8561 | 0.9594 |
91
- | 0.0811 | 12.22 | 5500 | 0.1842 | 0.8374 | 0.8872 | 0.8616 | 0.9613 |
92
- | 0.0735 | 13.33 | 6000 | 0.1858 | 0.8393 | 0.8777 | 0.8581 | 0.9616 |
93
- | 0.07 | 14.44 | 6500 | 0.1933 | 0.8349 | 0.8794 | 0.8566 | 0.9602 |
94
- | 0.0674 | 15.56 | 7000 | 0.1913 | 0.8384 | 0.8852 | 0.8612 | 0.9613 |
95
- | 0.0671 | 16.67 | 7500 | 0.1927 | 0.8340 | 0.8823 | 0.8575 | 0.9606 |
96
- | 0.0606 | 17.78 | 8000 | 0.1963 | 0.8398 | 0.8815 | 0.8601 | 0.9607 |
97
- | 0.0601 | 18.89 | 8500 | 0.1925 | 0.8395 | 0.8794 | 0.8590 | 0.9615 |
98
- | 0.0559 | 20.0 | 9000 | 0.1943 | 0.8392 | 0.8798 | 0.8590 | 0.9608 |
99
 
100
 
101
  ### Framework versions
 
24
  metrics:
25
  - name: Precision
26
  type: precision
27
+ value: 0.8408018867924528
28
  - name: Recall
29
  type: recall
30
+ value: 0.8835192069392813
31
  - name: F1
32
  type: f1
33
+ value: 0.8616314199395771
34
  - name: Accuracy
35
  type: accuracy
36
+ value: 0.9620919487995152
37
  ---
38
 
39
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
 
43
 
44
  This model is a fine-tuned version of [DeepPavlov/bert-base-bg-cs-pl-ru-cased](https://huggingface.co/DeepPavlov/bert-base-bg-cs-pl-ru-cased) on the cnec dataset.
45
  It achieves the following results on the evaluation set:
46
+ - Loss: 0.1622
47
+ - Precision: 0.8408
48
+ - Recall: 0.8835
49
+ - F1: 0.8616
50
+ - Accuracy: 0.9621
51
 
52
  ## Model description
53
 
 
72
  - seed: 42
73
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
74
  - lr_scheduler_type: linear
75
+ - num_epochs: 10
76
 
77
  ### Training results
78
 
79
  | Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
80
  |:-------------:|:-----:|:----:|:---------------:|:---------:|:------:|:------:|:--------:|
81
+ | 0.4517 | 1.11 | 500 | 0.2019 | 0.7361 | 0.7708 | 0.7530 | 0.9402 |
82
+ | 0.2103 | 2.22 | 1000 | 0.1727 | 0.7759 | 0.8381 | 0.8058 | 0.9518 |
83
+ | 0.1634 | 3.33 | 1500 | 0.1576 | 0.7943 | 0.8517 | 0.8220 | 0.9552 |
84
+ | 0.1359 | 4.44 | 2000 | 0.1661 | 0.8093 | 0.8625 | 0.8350 | 0.9567 |
85
+ | 0.1214 | 5.56 | 2500 | 0.1619 | 0.8237 | 0.8625 | 0.8426 | 0.9579 |
86
+ | 0.0986 | 6.67 | 3000 | 0.1605 | 0.8256 | 0.8761 | 0.8501 | 0.9606 |
87
+ | 0.0903 | 7.78 | 3500 | 0.1634 | 0.8343 | 0.8736 | 0.8535 | 0.9604 |
88
+ | 0.0805 | 8.89 | 4000 | 0.1611 | 0.8403 | 0.8823 | 0.8608 | 0.9614 |
89
+ | 0.0797 | 10.0 | 4500 | 0.1622 | 0.8408 | 0.8835 | 0.8616 | 0.9621 |
 
 
 
 
 
 
 
 
 
90
 
91
 
92
  ### Framework versions