amharic-hate-speech

This model is a fine-tuned version of uhhlt/am-roberta on an AmahricHateSpeechRANL dataset. It achieves the following results on the evaluation set:

• Loss: 0.6437
• Accuracy: 0.7373
• Precision: 0.7216
• Recall: 0.7149
• F1: 0.7180

How to use it

from transformers import pipeline
amhate_classifier = pipeline("text-classification", model="uhhlt/amharic-hate-speech")
amhate_classifier(["🌳☘️ 🌳☘️ለልጅ ልጅ የሚተላለፍ ዘመን ተሻጋሪ ኢንቨስትመንት !!!🌳☘️ 🌳☘️።", 
                  "አንተ አሁን ምን የሚሉህ ነህ? ግኡዝ አምላኪ ከመሆን ያድነን። ሰውን ያክል ፍጡር እየሞተ ለዛፍ ይሄን ያክል ማምለክ ጤነኝነት አይመስልም ። ፋኖ 100% ያሸንፋል",
                  "በአናትህ ተተከል ባንዳ ተላላኪ"])

Output

[{'label': 'normal', 'score': 0.8840981721878052},
 {'label': 'hate', 'score': 0.519339382648468},
 {'label': 'hate', 'score': 0.9630571007728577}]

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 128
• eval_batch_size: 128
• seed: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 8

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
0.8441	1.0	94	0.6699	0.7053	0.6913	0.6640	0.6737
0.6199	2.0	188	0.6505	0.72	0.7060	0.6995	0.6994
0.5295	3.0	282	0.6240	0.736	0.7201	0.7125	0.7159
0.4614	4.0	376	0.6437	0.7373	0.7216	0.7149	0.7180
0.3955	5.0	470	0.6922	0.7207	0.7001	0.7072	0.7031
0.3529	6.0	564	0.6995	0.7247	0.7050	0.7029	0.7039
0.3076	7.0	658	0.7352	0.7253	0.7067	0.7000	0.7031
0.2863	8.0	752	0.7470	0.7227	0.7019	0.6983	0.7000

Framework versions

• Transformers 4.32.1
• Pytorch 2.0.1+cu117
• Datasets 2.14.4
• Tokenizers 0.13.3