xlm-roberta-large-xnli-v5.0

This model is a fine-tuned version of joeddav/xlm-roberta-large-xnli on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4987
• F1 Macro: 0.8279
• F1 Micro: 0.8288
• Accuracy Balanced: 0.8278
• Accuracy: 0.8288
• Precision Macro: 0.8281
• Recall Macro: 0.8278
• Precision Micro: 0.8288
• Recall Micro: 0.8288

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: 9e-06
• train_batch_size: 8
• eval_batch_size: 64
• seed: 40
• gradient_accumulation_steps: 4
• 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: 0.06
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	F1 Macro	F1 Micro	Accuracy Balanced	Accuracy	Precision Macro	Recall Macro	Precision Micro	Recall Micro
0.3851	0.85	200	0.4586	0.8017	0.8025	0.8029	0.8025	0.8012	0.8029	0.8025	0.8025
0.2689	1.69	400	0.4498	0.8137	0.8147	0.8145	0.8147	0.8133	0.8145	0.8147	0.8147
0.194	2.54	600	0.5334	0.8244	0.8253	0.8252	0.8253	0.8239	0.8252	0.8253	0.8253

eval result

Datasets	asadfgglie/nli-zh-tw-all/test	asadfgglie/BanBan_2024-10-17-facial_expressions-nli/test	eval_dataset	test_dataset
eval_loss	0.535	0.278	0.552	0.499
eval_f1_macro	0.817	0.916	0.823	0.828
eval_f1_micro	0.818	0.916	0.824	0.829
eval_accuracy_balanced	0.817	0.917	0.823	0.828
eval_accuracy	0.818	0.916	0.824	0.829
eval_precision_macro	0.817	0.917	0.823	0.828
eval_recall_macro	0.817	0.917	0.823	0.828
eval_precision_micro	0.818	0.916	0.824	0.829
eval_recall_micro	0.818	0.916	0.824	0.829
eval_runtime	50.89	0.639	11.177	44.352
eval_samples_per_second	167.026	1480.253	169.012	170.387
eval_steps_per_second	2.613	23.471	2.684	2.683
Size of dataset	8500	946	1889	7557

Framework versions

• Transformers 4.33.3
• Pytorch 2.5.1+cu121
• Datasets 2.14.7
• Tokenizers 0.13.3