finetuned-ner

This model is a fine-tuned version of deepset/deberta-v3-base-squad2 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4783
• Precision: 0.3264
• Recall: 0.3591
• F1: 0.3420
• Accuracy: 0.8925

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 30
• mixed_precision_training: Native AMP
• label_smoothing_factor: 0.05

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
39.8167	1.0	760	0.3957	0.1844	0.2909	0.2257	0.8499
21.7333	2.0	1520	0.3853	0.2118	0.3273	0.2571	0.8546
13.8859	3.0	2280	0.3631	0.2443	0.2909	0.2656	0.8789
20.6586	4.0	3040	0.3961	0.2946	0.3455	0.3180	0.8753
13.8654	5.0	3800	0.3821	0.2791	0.3273	0.3013	0.8877
12.6942	6.0	4560	0.4393	0.3122	0.3364	0.3239	0.8909
25.0549	7.0	5320	0.4542	0.3106	0.3727	0.3388	0.8824
5.6816	8.0	6080	0.4432	0.2820	0.3409	0.3086	0.8774
13.1296	9.0	6840	0.4509	0.2884	0.35	0.3162	0.8824
7.7173	10.0	7600	0.4265	0.3170	0.3818	0.3464	0.8919
6.7922	11.0	8360	0.4749	0.3320	0.3818	0.3552	0.8892
5.4287	12.0	9120	0.4564	0.2917	0.3818	0.3307	0.8805
7.4153	13.0	9880	0.4735	0.2963	0.3273	0.3110	0.8871
9.1154	14.0	10640	0.4553	0.3416	0.3773	0.3585	0.8894
5.999	15.0	11400	0.4489	0.3203	0.4091	0.3593	0.8880
9.5128	16.0	12160	0.4947	0.3164	0.3682	0.3403	0.8883
5.6713	17.0	12920	0.4705	0.3527	0.3864	0.3688	0.8919
12.2119	18.0	13680	0.4617	0.3123	0.3591	0.3340	0.8857
8.5658	19.0	14440	0.4764	0.3092	0.35	0.3284	0.8944
11.0664	20.0	15200	0.4557	0.3187	0.3636	0.3397	0.8905
6.7161	21.0	15960	0.4468	0.3210	0.3955	0.3544	0.8956
9.0448	22.0	16720	0.5120	0.2872	0.3682	0.3227	0.8792
6.573	23.0	17480	0.4990	0.3307	0.3773	0.3524	0.8869
5.0543	24.0	18240	0.4763	0.3028	0.3455	0.3227	0.8899
6.8797	25.0	19000	0.4814	0.2780	0.3273	0.3006	0.8913
7.7544	26.0	19760	0.4695	0.3024	0.3409	0.3205	0.8946
4.8346	27.0	20520	0.4849	0.3154	0.3455	0.3297	0.8931
4.4766	28.0	21280	0.4809	0.2925	0.3364	0.3129	0.8913
7.9149	29.0	22040	0.4756	0.3238	0.3591	0.3405	0.8930
7.3033	30.0	22800	0.4783	0.3264	0.3591	0.3420	0.8925

Framework versions

• Transformers 4.25.1
• Pytorch 1.7.1
• Datasets 2.8.0
• Tokenizers 0.13.2