mDeBERTa-v3-base-xnli-multilingual-zeroshot-v1.2

This model is a fine-tuned version of 61347023S/mDeBERTa-v3-base-xnli-multilingual-zeroshot-v1.1 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7018
• F1 Macro: 0.8729
• F1 Micro: 0.8745
• Accuracy Balanced: 0.8716
• Accuracy: 0.8745
• Precision Macro: 0.8748
• Recall Macro: 0.8716
• Precision Micro: 0.8745
• Recall Micro: 0.8745

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

Datasets	asadfgglie/nli-zh-tw-all/test	asadfgglie/BanBan_2024-10-17-facial_expressions-nli	eval_dataset
Accuracy	0.875	0.953	0.871
Inference text/sec (RTX4090ti, batch=128)	126.0	829.0	125.0

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 128
• seed: 35
• gradient_accumulation_steps: 2
• 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
0.0962	0.17	200	0.5626	0.8709	0.8719	0.8714	0.8719	0.8704	0.8714	0.8719	0.8719
0.1218	0.34	400	0.5258	0.8635	0.8650	0.8629	0.8650	0.8643	0.8629	0.8650	0.8650
0.1246	0.51	600	0.4964	0.8652	0.8671	0.8635	0.8671	0.8679	0.8635	0.8671	0.8671
0.1225	0.68	800	0.5676	0.8618	0.8629	0.8623	0.8629	0.8614	0.8623	0.8629	0.8629
0.1429	0.85	1000	0.4402	0.8651	0.8666	0.8643	0.8666	0.8660	0.8643	0.8666	0.8666
0.1129	1.02	1200	0.5230	0.8688	0.8703	0.8679	0.8703	0.8699	0.8679	0.8703	0.8703
0.0921	1.19	1400	0.6435	0.8503	0.8534	0.8473	0.8534	0.8574	0.8473	0.8534	0.8534
0.0972	1.35	1600	0.5313	0.8635	0.8650	0.8628	0.8650	0.8644	0.8628	0.8650	0.8650
0.0883	1.52	1800	0.6088	0.8682	0.8692	0.8688	0.8692	0.8678	0.8688	0.8692	0.8692
0.0985	1.69	2000	0.5890	0.8696	0.8708	0.8693	0.8708	0.8698	0.8693	0.8708	0.8708
0.0838	1.86	2200	0.6647	0.8634	0.8650	0.8626	0.8650	0.8645	0.8626	0.8650	0.8650
0.0703	2.03	2400	0.6527	0.8712	0.8729	0.8699	0.8729	0.8732	0.8699	0.8729	0.8729
0.0639	2.2	2600	0.6665	0.8695	0.8714	0.8680	0.8714	0.8720	0.8680	0.8714	0.8714
0.059	2.37	2800	0.7361	0.8668	0.8687	0.8650	0.8687	0.8696	0.8650	0.8687	0.8687
0.062	2.54	3000	0.6719	0.8742	0.8756	0.8735	0.8756	0.8751	0.8735	0.8756	0.8756
0.0419	2.71	3200	0.7057	0.8734	0.8751	0.8722	0.8751	0.8753	0.8722	0.8751	0.8751
0.0539	2.88	3400	0.7020	0.8728	0.8745	0.8713	0.8745	0.8751	0.8713	0.8745	0.8745

Framework versions

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