cf6428a4a3a4ea0698b092729e5f62e5

This model is a fine-tuned version of studio-ousia/luke-japanese-large on the contemmcm/hate-speech-and-offensive-language dataset. It achieves the following results on the evaluation set:

• Loss: 0.6790
• Data Size: 1.0
• Epoch Runtime: 115.6747
• Accuracy: 0.7672
• F1 Macro: 0.2894

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Data Size	Epoch Runtime	Accuracy	F1 Macro
No log	0	0	1.0954	0	8.2836	0.2238	0.1679
No log	1	619	0.6991	0.0078	9.1900	0.7672	0.2894
No log	2	1238	0.6281	0.0156	11.0503	0.7672	0.2894
0.0148	3	1857	0.5711	0.0312	13.1430	0.8168	0.4696
0.0148	4	2476	0.4571	0.0625	16.7996	0.8295	0.5230
0.699	5	3095	0.6862	0.125	24.6249	0.7672	0.2894
0.0638	6	3714	0.6828	0.25	38.5782	0.7672	0.2894
0.6656	7	4333	0.6816	0.5	66.7274	0.7672	0.2894
0.6739	8.0	4952	0.6790	1.0	115.6747	0.7672	0.2894

Framework versions

• Transformers 4.57.0
• Pytorch 2.8.0+cu128
• Datasets 4.3.0
• Tokenizers 0.22.1