xtremedistil-l6-h384-uncased-v5.0

This model is a fine-tuned version of microsoft/xtremedistil-l6-h384-uncased on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5748
• F1 Macro: 0.6478
• F1 Micro: 0.6513
• Accuracy Balanced: 0.6608
• Accuracy: 0.6513
• Precision Macro: 0.6723
• Recall Macro: 0.6608
• Precision Micro: 0.6513
• Recall Micro: 0.6513

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: 40
• 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
0.6487	0.85	200	0.6075	0.5947	0.6188	0.6447	0.6188	0.7143	0.6447	0.6188	0.6188
0.6417	1.69	400	0.5900	0.6186	0.6257	0.6423	0.6257	0.6622	0.6423	0.6257	0.6257
0.6328	2.54	600	0.5822	0.6477	0.6485	0.6574	0.6485	0.6613	0.6574	0.6485	0.6485

eval result

Datasets	asadfgglie/nli-zh-tw-all/test	asadfgglie/BanBan_2024-10-17-facial_expressions-nli/test	eval_dataset	test_dataset
eval_loss	0.562	0.701	0.579	0.575
eval_f1_macro	0.666	0.478	0.645	0.648
eval_f1_micro	0.67	0.479	0.648	0.651
eval_accuracy_balanced	0.684	0.481	0.662	0.661
eval_accuracy	0.67	0.479	0.648	0.651
eval_precision_macro	0.702	0.48	0.674	0.672
eval_recall_macro	0.684	0.481	0.662	0.661
eval_precision_micro	0.67	0.479	0.648	0.651
eval_recall_micro	0.67	0.479	0.648	0.651
eval_runtime	9.241	0.208	1.943	7.696
eval_samples_per_second	919.801	4553.642	972.241	981.982
eval_steps_per_second	7.25	38.509	7.72	7.797
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