vihsd-xlmr-hate-speech

This model is a fine-tuned version of xlm-roberta-base on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4471
• Accuracy: 0.8496
• Precision Macro: 0.6512
• Recall Macro: 0.6691
• Macro F1: 0.6591
• Weighted F1: 0.8517
• Precision Clean: 0.9298
• Recall Clean: 0.9148
• F1 Clean: 0.9223
• Precision Offensive: 0.4804
• Recall Offensive: 0.4667
• F1 Offensive: 0.4734
• Precision Hate: 0.5434
• Recall Hate: 0.6259
• F1 Hate: 0.5818
• Critical F1: 0.5276
• Critical Recall: 0.5463
• Offensive Priority F1: 0.5059
• Offensive Priority Recall: 0.5144
• Balanced Critical F1: 0.5539
• Balanced Critical Recall: 0.5709

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: 16
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10.0
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Accuracy	Balanced Critical F1	Balanced Critical Recall	Critical F1	Critical Recall	F1 Clean	F1 Hate	F1 Offensive	Validation Loss	Macro F1	Offensive Priority F1	Offensive Priority Recall	Precision Clean	Precision Hate	Precision Offensive	Precision Macro	Recall Clean	Recall Hate	Recall Offensive	Recall Macro	Weighted F1
0.2894	0.9997	1770	0.7927	0.5005	0.597	0.4729	0.5796	0.8868	0.5614	0.3845	0.2184	0.6109	0.4376	0.5744	0.9394	0.5333	0.291	0.5879	0.8397	0.5926	0.5667	0.6663	0.8135
0.176	2.0	3541	0.7958	0.5117	0.6321	0.4849	0.6177	0.8877	0.5253	0.4444	0.2149	0.6192	0.4687	0.5878	0.9498	0.4231	0.3762	0.5831	0.8333	0.6926	0.5429	0.6896	0.8154
0.1165	2.9997	5311	0.8041	0.5081	0.6	0.4805	0.582	0.8935	0.5266	0.4344	0.2400	0.6182	0.4621	0.5511	0.9402	0.4384	0.3813	0.5866	0.8513	0.6593	0.5048	0.6718	0.8195
0.0744	4.0	7082	0.8515	0.5498	0.5531	0.5232	0.5267	0.9232	0.6019	0.4444	0.2723	0.6565	0.4917	0.4989	0.9247	0.6075	0.4324	0.6549	0.9217	0.5963	0.4571	0.6584	0.8522
0.0537	4.9997	8852	0.8462	0.537	0.5572	0.5095	0.5317	0.9215	0.5898	0.4293	0.3155	0.6469	0.4774	0.4867	0.9302	0.5437	0.44	0.638	0.9129	0.6444	0.419	0.6588	0.8484
0.0387	6.0	10623	0.8583	0.5401	0.5328	0.5123	0.5042	0.9292	0.5972	0.4274	0.3731	0.6513	0.4783	0.4511	0.9252	0.5621	0.5032	0.6635	0.9333	0.637	0.3714	0.6473	0.8553
0.0282	6.9997	12393	0.848	0.5345	0.5394	0.5068	0.5122	0.9218	0.5814	0.4322	0.4123	0.6451	0.4769	0.4711	0.9237	0.5515	0.4574	0.6442	0.9199	0.6148	0.4095	0.6481	0.848
0.0253	7.9997	14160	0.8496	0.5539	0.5709	0.5276	0.5463	0.9223	0.5818	0.4734	0.4471	0.6591	0.5059	0.5144	0.9298	0.5434	0.4804	0.6512	0.9148	0.6259	0.4667	0.6691	0.8517
0.0169	9.0	15931	0.848	0.536	0.5391	0.5085	0.5119	0.9216	0.5796	0.4373	0.4845	0.6462	0.48	0.4767	0.9229	0.5606	0.4518	0.6451	0.9203	0.6	0.4238	0.648	0.8481
0.0156	9.9997	17700	0.5180	0.8545	0.6586	0.6529	0.6546	0.8531	0.9234	0.9268	0.9251	0.4751	0.4095	0.4399	0.5773	0.6222	0.5989	0.5194	0.5159	0.4876	0.4733	0.5465	0.5433

Framework versions

• Transformers 4.40.2
• Pytorch 2.11.0+cu128
• Datasets 2.19.2
• Tokenizers 0.19.1