xlm-roberta-large-xnli-anli-v5.0

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

• Loss: 0.5120
• F1 Macro: 0.8215
• F1 Micro: 0.8223
• Accuracy Balanced: 0.8216
• Accuracy: 0.8223
• Precision Macro: 0.8215
• Recall Macro: 0.8216
• Precision Micro: 0.8223
• Recall Micro: 0.8223

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
0.3779	0.85	200	0.4494	0.8020	0.8020	0.8084	0.8020	0.8088	0.8084	0.8020	0.8020
0.2646	1.69	400	0.4425	0.8113	0.8121	0.8126	0.8121	0.8108	0.8126	0.8121	0.8121
0.1961	2.54	600	0.5222	0.8131	0.8147	0.8129	0.8147	0.8135	0.8129	0.8147	0.8147

eval result

Datasets	asadfgglie/nli-zh-tw-all/test	asadfgglie/BanBan_2024-10-17-facial_expressions-nli/test	eval_dataset	test_dataset
eval_loss	0.541	0.26	0.517	0.512
eval_f1_macro	0.809	0.918	0.814	0.822
eval_f1_micro	0.81	0.918	0.815	0.822
eval_accuracy_balanced	0.809	0.918	0.815	0.822
eval_accuracy	0.81	0.918	0.815	0.822
eval_precision_macro	0.809	0.918	0.814	0.821
eval_recall_macro	0.809	0.918	0.815	0.822
eval_precision_micro	0.81	0.918	0.815	0.822
eval_recall_micro	0.81	0.918	0.815	0.822
eval_runtime	50.676	0.617	11.139	44.314
eval_samples_per_second	167.732	1532.353	169.579	170.532
eval_steps_per_second	2.625	24.297	2.693	2.685
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