Structured-FP16-KD-NID

This model is a fine-tuned version of huawei-noah/TinyBERT_General_4L_312D on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0343
• Accuracy: 0.9926
• Precision: 0.9546
• Recall: 0.9514
• F1 score: 0.9515

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: 5e-05
• train_batch_size: 650
• eval_batch_size: 650
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1 score
0.1509	1.0	1828	0.1228	0.9830	0.8658	0.8392	0.8343
0.0961	2.0	3656	0.0884	0.9865	0.9110	0.8672	0.8652
0.0812	3.0	5484	0.0763	0.9879	0.9152	0.8869	0.8882
0.073	4.0	7312	0.0650	0.9889	0.9252	0.8966	0.9007
0.0605	5.0	9140	0.0568	0.9897	0.9272	0.9026	0.9075
0.0571	6.0	10968	0.0516	0.9902	0.9131	0.9202	0.9156
0.0487	7.0	12796	0.0460	0.9909	0.9282	0.9228	0.9247
0.045	8.0	14624	0.0471	0.9907	0.9219	0.9208	0.9196
0.0396	9.0	16452	0.0443	0.9910	0.9279	0.9253	0.9258
0.0409	10.0	18280	0.0422	0.9913	0.9269	0.9315	0.9288
0.0366	11.0	20108	0.0397	0.9916	0.9264	0.9359	0.9308
0.037	12.0	21936	0.0387	0.9919	0.9336	0.9307	0.9308
0.0367	13.0	23764	0.0374	0.9921	0.9317	0.9340	0.9320
0.0315	14.0	25592	0.0379	0.9921	0.9313	0.9376	0.9338
0.0325	15.0	27420	0.0353	0.9925	0.9319	0.9389	0.9349
0.0299	16.0	29248	0.0351	0.9924	0.9324	0.9376	0.9347
0.028	17.0	31076	0.0350	0.9924	0.9426	0.9462	0.9424
0.0303	18.0	32904	0.0347	0.9926	0.9541	0.9494	0.9501
0.0255	19.0	34732	0.0345	0.9926	0.9520	0.9501	0.9501
0.0259	20.0	36560	0.0343	0.9926	0.9546	0.9514	0.9515

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1