xlm-roberta-large-xnli-anli-v2.0

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: 0.3375
• F1 Macro: 0.8802
• F1 Micro: 0.8809
• Accuracy Balanced: 0.8798
• Accuracy: 0.8809
• Precision Macro: 0.8808
• Recall Macro: 0.8798
• Precision Micro: 0.8809
• Recall Micro: 0.8809

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
0.4942	0.17	200	0.4413	0.8081	0.8089	0.8093	0.8089	0.8076	0.8093	0.8089	0.8089
0.4114	0.34	400	0.3991	0.8227	0.8232	0.8250	0.8232	0.8226	0.8250	0.8232	0.8232
0.3467	0.51	600	0.3584	0.8388	0.8391	0.8421	0.8391	0.8396	0.8421	0.8391	0.8391
0.3402	0.68	800	0.3620	0.8534	0.8544	0.8536	0.8544	0.8532	0.8536	0.8544	0.8544
0.3304	0.85	1000	0.3385	0.8566	0.8576	0.8567	0.8576	0.8565	0.8567	0.8576	0.8576
0.3234	1.02	1200	0.3456	0.8637	0.8650	0.8631	0.8650	0.8645	0.8631	0.8650	0.8650
0.2702	1.19	1400	0.3201	0.8606	0.8613	0.8616	0.8613	0.8600	0.8616	0.8613	0.8613
0.2581	1.36	1600	0.3233	0.8619	0.8624	0.8639	0.8624	0.8615	0.8639	0.8624	0.8624
0.2414	1.52	1800	0.3451	0.8674	0.8687	0.8664	0.8687	0.8687	0.8664	0.8687	0.8687
0.2687	1.69	2000	0.3415	0.8577	0.8608	0.8544	0.8608	0.8677	0.8544	0.8608	0.8608
0.2518	1.86	2200	0.3378	0.8684	0.8692	0.8688	0.8692	0.8681	0.8688	0.8692	0.8692
0.2182	2.03	2400	0.3581	0.8698	0.8708	0.8697	0.8708	0.8700	0.8697	0.8708	0.8708
0.1919	2.2	2600	0.3671	0.8677	0.8687	0.8676	0.8687	0.8678	0.8676	0.8687	0.8687
0.1771	2.37	2800	0.3790	0.8709	0.8719	0.8707	0.8719	0.8710	0.8707	0.8719	0.8719
0.1793	2.54	3000	0.3856	0.8687	0.8692	0.8701	0.8692	0.8680	0.8701	0.8692	0.8692
0.1909	2.71	3200	0.3777	0.8686	0.8698	0.8682	0.8698	0.8691	0.8682	0.8698	0.8698
0.2021	2.88	3400	0.3685	0.8701	0.8708	0.8710	0.8708	0.8696	0.8710	0.8708	0.8708

eval result

Datasets	asadfgglie/nli-zh-tw-all/test	asadfgglie/BanBan_2024-10-17-facial_expressions-nli/test	eval_dataset	test_dataset
eval_loss	0.355	0.246	0.369	0.337
eval_f1_macro	0.872	0.932	0.872	0.88
eval_f1_micro	0.873	0.932	0.873	0.881
eval_accuracy_balanced	0.872	0.932	0.873	0.88
eval_accuracy	0.873	0.932	0.873	0.881
eval_precision_macro	0.873	0.932	0.872	0.881
eval_recall_macro	0.872	0.932	0.873	0.88
eval_precision_micro	0.873	0.932	0.873	0.881
eval_recall_micro	0.873	0.932	0.873	0.881
eval_runtime	50.57	0.614	11.127	44.11
eval_samples_per_second	168.083	1541.862	169.759	171.323
eval_steps_per_second	2.63	24.448	2.696	2.698
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