mDeBERTa-v3-base-xnli-multilingual-zeroshot-v5.0-nli-downsample-and-non-nli

This model is merge dataset stratege version of v3.0 and v4.0.

This model is a fine-tuned version of MoritzLaurer/mDeBERTa-v3-base-xnli-multilingual-nli-2mil7 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4531
• F1 Macro: 0.8330
• F1 Micro: 0.8337
• Accuracy Balanced: 0.8331
• Accuracy: 0.8337
• Precision Macro: 0.8330
• Recall Macro: 0.8331
• Precision Micro: 0.8337
• Recall Micro: 0.8337

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 128
• seed: 20241201
• 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
0.3748	0.85	200	0.4218	0.7971	0.7999	0.7970	0.7999	0.7973	0.7970	0.7999	0.7999
0.2693	1.69	400	0.4523	0.8061	0.8078	0.8077	0.8078	0.8053	0.8077	0.8078	0.8078
0.1905	2.54	600	0.4720	0.8226	0.8242	0.8241	0.8242	0.8217	0.8241	0.8242	0.8242

Eval results

Datasets	asadfgglie/nli-zh-tw-all/test	asadfgglie/BanBan_2024-10-17-facial_expressions-nli/test	eval_dataset	test_dataset
eval_loss	0.48	0.269	0.484	0.453
eval_f1_macro	0.821	0.909	0.816	0.833
eval_f1_micro	0.822	0.909	0.818	0.834
eval_accuracy_balanced	0.821	0.909	0.816	0.833
eval_accuracy	0.822	0.909	0.818	0.834
eval_precision_macro	0.821	0.909	0.816	0.833
eval_recall_macro	0.821	0.909	0.816	0.833
eval_precision_micro	0.822	0.909	0.818	0.834
eval_recall_micro	0.822	0.909	0.818	0.834
eval_runtime	239.87	4.066	58.954	236.797
eval_samples_per_second	35.436	232.633	32.042	31.913
eval_steps_per_second	0.279	1.967	0.254	0.253
epoch	2.99	2.99	2.99	2.99
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