Hugging Face's logo

thejosango
/
nuha-binary

Text Classification
Transformers
PyTorch
Arabic
bert
hate-speech
gender-based-violence
arabic
binary-classification
pilot
Eval Results (legacy)
text-embeddings-inference
Model card Files
xet
 Community
yazansh commited on
Commit
ee0cc92
·
verified ·
1 Parent(s): 714ad98

binary-27

Browse files
Files changed (4) hide show
  1. README.md +24 -40
  2. config.toml +9 -9
  3. pytorch_model.bin +1 -1
  4. training_args.bin +1 -1
README.md CHANGED
@@ -23,13 +23,13 @@ model-index:
23
  metrics:
24
  - name: F1
25
  type: f1
26
- value: 0.6709883502442691
27
  - name: Precision
28
  type: precision
29
- value: 0.6826610590709233
30
  - name: Recall
31
  type: recall
32
- value: 0.6597081101053021
33
  ---
34
 
35
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
@@ -39,10 +39,10 @@ should probably proofread and complete it, then remove this comment. -->
39
 
40
  This model is a fine-tuned version of [thejosango/nuha-mlm](https://huggingface.co/thejosango/nuha-mlm) on the nuha-dataset dataset.
41
  It achieves the following results on the evaluation set:
42
- - Loss: 0.5203
43
- - F1: 0.6710
44
- - Precision: 0.6827
45
- - Recall: 0.6597
46
  - Support: None
47
 
48
  ## Model description
@@ -63,47 +63,31 @@ More information needed
63
 
64
  The following hyperparameters were used during training:
65
  - learning_rate: 1e-05
66
- - train_batch_size: 32
67
- - eval_batch_size: 32
68
  - seed: 42
69
  - gradient_accumulation_steps: 2
70
- - total_train_batch_size: 64
71
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
72
- - lr_scheduler_type: linear
73
  - lr_scheduler_warmup_steps: 1000.0
74
- - num_epochs: 20
75
  - label_smoothing_factor: 0.1
76
 
77
  ### Training results
78
 
79
- | Training Loss | Epoch | Step | Validation Loss | F1 | Precision | Recall | Support |
80
- |:-------------:|:-----:|:-----:|:---------------:|:------:|:---------:|:------:|:-------:|
81
- | 1.2252 | 0.23 | 500 | 0.6446 | 0.5213 | 0.6006 | 0.4606 | None |
82
- | 0.8851 | 0.46 | 1000 | 0.5788 | 0.6011 | 0.6123 | 0.5904 | None |
83
- | 0.7632 | 0.7 | 1500 | 0.5560 | 0.5957 | 0.6485 | 0.5509 | None |
84
- | 0.684 | 0.93 | 2000 | 0.5481 | 0.6076 | 0.6506 | 0.5699 | None |
85
- | 0.6334 | 1.16 | 2500 | 0.6265 | 0.6462 | 0.5199 | 0.8535 | None |
86
- | 0.6137 | 1.39 | 3000 | 0.5378 | 0.5593 | 0.7224 | 0.4563 | None |
87
- | 0.609 | 1.63 | 3500 | 0.5348 | 0.6486 | 0.6499 | 0.6473 | None |
88
- | 0.5756 | 1.86 | 4000 | 0.5272 | 0.6182 | 0.6966 | 0.5557 | None |
89
- | 0.5667 | 2.09 | 4500 | 0.5257 | 0.6035 | 0.7194 | 0.5197 | None |
90
- | 0.5551 | 2.32 | 5000 | 0.5193 | 0.6456 | 0.6837 | 0.6115 | None |
91
- | 0.5549 | 2.56 | 5500 | 0.5173 | 0.6236 | 0.7191 | 0.5505 | None |
92
- | 0.5507 | 2.79 | 6000 | 0.5279 | 0.6675 | 0.6542 | 0.6813 | None |
93
- | 0.5505 | 3.02 | 6500 | 0.5164 | 0.6534 | 0.6916 | 0.6192 | None |
94
- | 0.5298 | 3.25 | 7000 | 0.5232 | 0.6687 | 0.6628 | 0.6747 | None |
95
- | 0.5313 | 3.49 | 7500 | 0.5128 | 0.6603 | 0.6934 | 0.6303 | None |
96
- | 0.5209 | 3.72 | 8000 | 0.5285 | 0.6800 | 0.6513 | 0.7113 | None |
97
- | 0.5214 | 3.95 | 8500 | 0.5127 | 0.6443 | 0.7128 | 0.5878 | None |
98
- | 0.5033 | 4.18 | 9000 | 0.5179 | 0.6341 | 0.7268 | 0.5623 | None |
99
- | 0.5055 | 4.41 | 9500 | 0.5214 | 0.6239 | 0.7347 | 0.5422 | None |
100
- | 0.5013 | 4.65 | 10000 | 0.5230 | 0.6794 | 0.6626 | 0.6972 | None |
101
- | 0.5107 | 4.88 | 10500 | 0.5127 | 0.6656 | 0.7012 | 0.6335 | None |
102
- | 0.4862 | 5.11 | 11000 | 0.5447 | 0.5848 | 0.7670 | 0.4726 | None |
103
- | 0.4814 | 5.34 | 11500 | 0.5216 | 0.6217 | 0.7386 | 0.5367 | None |
104
- | 0.4877 | 5.58 | 12000 | 0.5176 | 0.6375 | 0.7255 | 0.5684 | None |
105
- | 0.4893 | 5.81 | 12500 | 0.5180 | 0.6463 | 0.7129 | 0.5912 | None |
106
- | 0.4779 | 6.04 | 13000 | 0.5203 | 0.6710 | 0.6827 | 0.6597 | None |
107
 
108
 
109
  ### Framework versions
 
23
  metrics:
24
  - name: F1
25
  type: f1
26
+ value: 0.36679201619901647
27
  - name: Precision
28
  type: precision
29
+ value: 0.8447701532311792
30
  - name: Recall
31
  type: recall
32
+ value: 0.23425087751708848
33
  ---
34
 
35
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
 
39
 
40
  This model is a fine-tuned version of [thejosango/nuha-mlm](https://huggingface.co/thejosango/nuha-mlm) on the nuha-dataset dataset.
41
  It achieves the following results on the evaluation set:
42
+ - Loss: 0.4531
43
+ - F1: 0.3668
44
+ - Precision: 0.8448
45
+ - Recall: 0.2343
46
  - Support: None
47
 
48
  ## Model description
 
63
 
64
  The following hyperparameters were used during training:
65
  - learning_rate: 1e-05
66
+ - train_batch_size: 8
67
+ - eval_batch_size: 8
68
  - seed: 42
69
  - gradient_accumulation_steps: 2
70
+ - total_train_batch_size: 16
71
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
72
+ - lr_scheduler_type: constant
73
  - lr_scheduler_warmup_steps: 1000.0
74
+ - num_epochs: 30
75
  - label_smoothing_factor: 0.1
76
 
77
  ### Training results
78
 
79
+ | Training Loss | Epoch | Step | Validation Loss | F1 | Precision | Recall | Support |
80
+ |:-------------:|:-----:|:----:|:---------------:|:------:|:---------:|:------:|:-------:|
81
+ | 0.8428 | 0.06 | 500 | 0.5318 | 0.4246 | 0.6560 | 0.3139 | None |
82
+ | 0.6948 | 0.13 | 1000 | 0.4850 | 0.4097 | 0.7445 | 0.2827 | None |
83
+ | 0.684 | 0.19 | 1500 | 0.4674 | 0.3861 | 0.7563 | 0.2592 | None |
84
+ | 0.5896 | 0.25 | 2000 | 0.4870 | 0.2807 | 0.8451 | 0.1683 | None |
85
+ | 0.5585 | 0.31 | 2500 | 0.4591 | 0.5076 | 0.7354 | 0.3876 | None |
86
+ | 0.5371 | 0.38 | 3000 | 0.4484 | 0.4326 | 0.7993 | 0.2965 | None |
87
+ | 0.5286 | 0.44 | 3500 | 0.4479 | 0.4129 | 0.8212 | 0.2758 | None |
88
+ | 0.5071 | 0.5 | 4000 | 0.4433 | 0.4647 | 0.7822 | 0.3305 | None |
89
+ | 0.5043 | 0.56 | 4500 | 0.4799 | 0.2584 | 0.8539 | 0.1522 | None |
90
+ | 0.5149 | 0.63 | 5000 | 0.4531 | 0.3668 | 0.8448 | 0.2343 | None |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
91
 
92
 
93
  ### Framework versions
config.toml CHANGED
@@ -1,13 +1,13 @@
1
  [experiment]
2
- name = "binary-26"
3
  type = "binary"
4
 
5
 
6
  [dataset]
7
  path = "thejosango/nuha-dataset"
8
  dataset_revision = "main"
9
- augment_ratio = 0.25
10
- undersampling_strategy = 0
11
 
12
 
13
  [model]
@@ -16,15 +16,15 @@ revision = "2caf9ebc5b275737c95f8bb16953288107a7131c"
16
 
17
 
18
  [training]
19
- num_train_epochs = 20
20
  warmup_steps = 1e3
21
- lr_scheduler_type = "linear"
22
  learning_rate = 1e-5
23
- per_device_train_batch_size = 32
24
- per_device_eval_batch_size = 32
25
  gradient_accumulation_steps = 2
26
  weight_decay = 0.01
27
  label_smoothing_factor = 0.1
28
- weighted_loss = false
29
- early_stopping_patience = 10
30
  early_stopping_threshold = 0.005
 
1
  [experiment]
2
+ name = "binary-27"
3
  type = "binary"
4
 
5
 
6
  [dataset]
7
  path = "thejosango/nuha-dataset"
8
  dataset_revision = "main"
9
+ augment_ratio = 0.0
10
+ undersampling_strategy = false
11
 
12
 
13
  [model]
 
16
 
17
 
18
  [training]
19
+ num_train_epochs = 30
20
  warmup_steps = 1e3
21
+ lr_scheduler_type = "constant"
22
  learning_rate = 1e-5
23
+ per_device_train_batch_size = 8
24
+ per_device_eval_batch_size = 8
25
  gradient_accumulation_steps = 2
26
  weight_decay = 0.01
27
  label_smoothing_factor = 0.1
28
+ weighted_loss = true
29
+ early_stopping_patience = 5
30
  early_stopping_threshold = 0.005
pytorch_model.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6274d0bde584699734a00b2cd752b54dcaf69e5a4753cab991977f54ca510145
3
  size 313992076
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:587e8391854237b42317246fd0b723aee9b71b6d155d6fa2f5f78109242cf9f9
3
  size 313992076
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0cc61a0a4012cb94dd109030c3f0bc5d60b87d98e45eac45d39704943b4e479e
3
  size 4091
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:12ec62d67b80d950a507d49d7653460381b8a047dc923e1e97dc8a60d3a0d858
3
  size 4091