xlm-roberta-large-xnli-v2.0

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: 0.3413
• F1 Macro: 0.8779
• F1 Micro: 0.8787
• Accuracy Balanced: 0.8773
• Accuracy: 0.8787
• Precision Macro: 0.8788
• Recall Macro: 0.8773
• Precision Micro: 0.8787
• Recall Micro: 0.8787

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
0.4814	0.17	200	0.4554	0.7851	0.7867	0.7852	0.7867	0.7850	0.7852	0.7867	0.7867
0.4031	0.34	400	0.4020	0.8228	0.8237	0.8235	0.8237	0.8223	0.8235	0.8237	0.8237
0.3425	0.51	600	0.3603	0.8450	0.8454	0.8473	0.8454	0.8448	0.8473	0.8454	0.8454
0.3374	0.68	800	0.3520	0.8518	0.8523	0.8538	0.8523	0.8514	0.8538	0.8523	0.8523
0.326	0.85	1000	0.3386	0.8529	0.8544	0.8521	0.8544	0.8541	0.8521	0.8544	0.8544
0.3059	1.02	1200	0.3425	0.8643	0.8650	0.8651	0.8650	0.8637	0.8651	0.8650	0.8650
0.2563	1.19	1400	0.3234	0.8708	0.8719	0.8703	0.8719	0.8713	0.8703	0.8719	0.8719
0.252	1.36	1600	0.3487	0.8580	0.8581	0.8616	0.8581	0.8590	0.8616	0.8581	0.8581
0.2323	1.52	1800	0.3576	0.8648	0.8666	0.8630	0.8666	0.8681	0.8630	0.8666	0.8666
0.2669	1.69	2000	0.3888	0.8461	0.8502	0.8425	0.8502	0.8603	0.8425	0.8502	0.8502
0.2514	1.86	2200	0.3323	0.8742	0.8751	0.8743	0.8751	0.8740	0.8743	0.8751	0.8751
0.1999	2.03	2400	0.3649	0.8759	0.8767	0.8762	0.8767	0.8755	0.8762	0.8767	0.8767
0.1764	2.2	2600	0.3889	0.8695	0.8708	0.8685	0.8708	0.8709	0.8685	0.8708	0.8708
0.1729	2.37	2800	0.3741	0.8676	0.8687	0.8674	0.8687	0.8679	0.8674	0.8687	0.8687
0.159	2.54	3000	0.3844	0.8760	0.8767	0.8772	0.8767	0.8754	0.8772	0.8767	0.8767
0.178	2.71	3200	0.3771	0.8693	0.8708	0.8680	0.8708	0.8714	0.8680	0.8708	0.8708
0.1893	2.88	3400	0.3678	0.8722	0.8729	0.8730	0.8729	0.8717	0.8730	0.8729	0.8729

eval result

Datasets	asadfgglie/nli-zh-tw-all/test	asadfgglie/BanBan_2024-10-17-facial_expressions-nli/test	eval_dataset	test_dataset
eval_loss	0.357	0.261	0.369	0.341
eval_f1_macro	0.872	0.919	0.874	0.878
eval_f1_micro	0.874	0.919	0.875	0.879
eval_accuracy_balanced	0.872	0.919	0.874	0.877
eval_accuracy	0.874	0.919	0.875	0.879
eval_precision_macro	0.873	0.919	0.874	0.879
eval_recall_macro	0.872	0.919	0.874	0.877
eval_precision_micro	0.874	0.919	0.875	0.879
eval_recall_micro	0.874	0.919	0.875	0.879
eval_runtime	50.977	0.625	11.165	44.322
eval_samples_per_second	166.741	1514.715	169.192	170.501
eval_steps_per_second	2.609	24.018	2.687	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