turkish_hate_speech

This model is a fine-tuned version of google-bert/bert-base-multilingual-cased on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3413
• Accuracy: 0.8566
• F1 Macro: 0.8564
• Precision Macro: 0.8612
• Recall Macro: 0.8580
• F1 Nefret: 0.8512
• F1 Hicbiri: 0.8615

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use paged_adamw_8bit with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.01
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1 Macro	Precision Macro	Recall Macro	F1 Nefret	F1 Hicbiri
0.5934	1.0	800	0.4766	0.7694	0.7670	0.7818	0.7697	0.7432	0.7907
0.4308	2.0	1600	0.3973	0.8184	0.8184	0.8186	0.8184	0.8212	0.8156
0.3898	3.0	2400	0.3548	0.8431	0.8430	0.8440	0.8432	0.8396	0.8465
0.3393	4.0	3200	0.3355	0.8538	0.8535	0.8566	0.8539	0.8474	0.8596
0.319	5.0	4000	0.3220	0.86	0.8600	0.8601	0.8600	0.8590	0.8610
0.3053	6.0	4800	0.3201	0.8641	0.8640	0.8654	0.8642	0.8603	0.8677
0.2887	7.0	5600	0.3166	0.8638	0.8634	0.8673	0.8639	0.8570	0.8699
0.2908	8.0	6400	0.3271	0.8634	0.8631	0.8679	0.8636	0.8559	0.8702
0.2764	9.0	7200	0.3207	0.8659	0.8657	0.8692	0.8661	0.8597	0.8717
0.2754	10.0	8000	0.3207	0.8656	0.8654	0.8689	0.8658	0.8594	0.8713

Framework versions

• PEFT 0.15.1
• Transformers 4.50.3
• Pytorch 2.5.1+cu121
• Datasets 3.5.0
• Tokenizers 0.21.0