mdeberta-v3-base-conll2003-en

This model is a fine-tuned version of microsoft/mdeberta-v3-base on the eriktks/conll2003 dataset (English split of the CONLL 2003). It achieves the following results on the evaluation set:

• Loss: 0.0342
• Precision: 0.9566
• Recall: 0.9650
• F1: 0.9608
• Accuracy: 0.9929

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 32
• seed: 42
• 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
• num_epochs: 5.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	1.0	439	0.0509	0.9303	0.9456	0.9379	0.9890
0.1482	2.0	878	0.0359	0.9501	0.9583	0.9542	0.9918
0.0335	3.0	1317	0.0338	0.9530	0.9615	0.9572	0.9924
0.0191	4.0	1756	0.0346	0.9538	0.9635	0.9586	0.9926
0.0137	5.0	2195	0.0342	0.9566	0.9650	0.9608	0.9929

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1