Hugging Face's logo

Cyrile
/
EuroBERT-210m-Quality-CL

Text Classification
Transformers
Safetensors
French
English
Spanish
eurobert
java
python
javascript
C/C++
custom_code
Model card Files
xet
 Community
Cyrile commited on
Commit
c5fe49a
·
verified ·
1 Parent(s): 92fef32

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +16 -16
README.md CHANGED
@@ -41,34 +41,34 @@ We compare two distinct approaches:
41
 
42
  | Catégorie | Global (NL + CL) | NL | CL |
43
  |:------------:|:----------------:|:-------------:|:-------------:|
44
- | **Harmfull** | 0.81 | 0.87 | 0.75 |
45
- | **Low** | 0.60 | 0.72 | 0.44 |
46
- | **Medium** | 0.60 | 0.74 | 0.49 |
47
- | **High** | 0.74 | 0.77 | 0.72 |
48
- | **Accuracy** | **0.70** | **0.78** | **0.62** |
49
 
50
 
51
  - **f1-score: Separate Models**
52
 
53
  | Catégorie | Global (NL + CL) | NL | CL |
54
  |:------------:|:----------------:|:-------------:|:-------------:|
55
- | **Harmfull** | 0.83 | 0.89 | 0.78 |
56
- | **Low** | 0.59 | 0.71 | 0.46 |
57
- | **Medium** | 0.63 | 0.77 | 0.49 |
58
- | **High** | 0.76 | 0.79 | 0.73 |
59
- | **Accuracy** | **0.71** | **0.80** | **0.63** |
60
 
61
 
62
  ## Key Performance Metrics:
63
  - **Unified Model (NL + CL)**:
64
- - Overall accuracy: ~69%
65
- - High reliability on harmful data (f1-score: 0.81)
66
 
67
  - **Separate Models**:
68
- - **Natural Language (NL)**: ~79% accuracy
69
- - Excellent performance on harmful data (f1-score: 0.89)
70
- - **Code Language (CL)**: ~63% accuracy
71
- - Good detection of harmful data (f1-score: 0.78)
72
 
73
  ## Training Dataset:
74
  - Public dataset available: [TempestTeam/dataset-quality](https://huggingface.co/datasets/TempestTeam/dataset-quality)
 
41
 
42
  | Catégorie | Global (NL + CL) | NL | CL |
43
  |:------------:|:----------------:|:-------------:|:-------------:|
44
+ | **Harmfull** | 0.86 | 0.93 | 0.79 |
45
+ | **Low** | 0.62 | 0.81 | 0.40 |
46
+ | **Medium** | 0.63 | 0.78 | 0.50 |
47
+ | **High** | 0.77 | 0.81 | 0.74 |
48
+ | **Accuracy** | **0.73** | **0.83** | **0.62** |
49
 
50
 
51
  - **f1-score: Separate Models**
52
 
53
  | Catégorie | Global (NL + CL) | NL | CL |
54
  |:------------:|:----------------:|:-------------:|:-------------:|
55
+ | **Harmfull** | 0.83 | 0.93 | 0.72 |
56
+ | **Low** | 0.64 | 0.76 | 0.53 |
57
+ | **Medium** | 0.63 | 0.76 | 0.52 |
58
+ | **High** | 0.79 | 0.81 | 0.76 |
59
+ | **Accuracy** | **0.73** | **0.82** | **0.63** |
60
 
61
 
62
  ## Key Performance Metrics:
63
  - **Unified Model (NL + CL)**:
64
+ - Overall accuracy: ~73%
65
+ - High reliability on harmful data (f1-score: 0.86)
66
 
67
  - **Separate Models**:
68
+ - **Natural Language (NL)**: ~82% accuracy
69
+ - Excellent performance on harmful data (f1-score: 0.93)
70
+ - **Code Language (CL)**: ~63% accuracy
71
+ - Good detection of harmful data (f1-score: 0.72)
72
 
73
  ## Training Dataset:
74
  - Public dataset available: [TempestTeam/dataset-quality](https://huggingface.co/datasets/TempestTeam/dataset-quality)