w2v-bert-2.0-lg-cv-5hr-v1

This model is a fine-tuned version of facebook/w2v-bert-2.0 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.8566
• Model Preparation Time: 0.0165
• Wer: 0.9775
• Cer: 0.8923

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: 16
• eval_batch_size: 8
• 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: cosine
• lr_scheduler_warmup_ratio: 0.01
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Cer
9.9607	0.9948	95	6.8754	0.0165	1.0	1.0
5.2586	2.0	191	4.0569	0.0165	1.0	1.0
3.4197	2.9948	286	3.0508	0.0165	1.0	1.0
2.9792	4.0	382	2.9586	0.0165	1.0	1.0
2.9646	4.9948	477	2.9354	0.0165	1.0	1.0
2.9169	6.0	573	2.9220	0.0165	1.0	1.0
2.9372	6.9948	668	2.9116	0.0165	1.0	1.0
2.8971	8.0	764	2.8998	0.0165	1.0	0.9811
2.918	8.9948	859	2.8893	0.0165	0.9983	0.9652
2.8795	10.0	955	2.8804	0.0165	0.9985	0.9534
2.9006	10.9948	1050	2.8683	0.0165	1.0	0.9048
2.8598	12.0	1146	2.8554	0.0165	1.0	0.9067
2.8776	12.9948	1241	2.8417	0.0165	1.0	0.8954
2.8393	14.0	1337	2.8407	0.0165	0.9970	0.9074
2.8637	14.9948	1432	2.8304	0.0165	0.9787	0.8824
2.8264	16.0	1528	2.8257	0.0165	0.9776	0.8934
2.846	16.9948	1623	2.8045	0.0165	1.0	0.8653
2.8001	18.0	1719	2.7907	0.0165	1.0022	0.8459
2.8103	18.9948	1814	2.7686	0.0165	0.9991	0.8579
2.7683	20.0	1910	2.7518	0.0165	0.9991	0.8534
2.7903	20.9948	2005	2.7481	0.0165	0.9980	0.8568
2.7561	22.0	2101	2.7468	0.0165	0.9991	0.8478
2.782	22.9948	2196	2.7383	0.0165	0.9978	0.8497
2.7473	24.0	2292	2.7345	0.0165	0.9993	0.8492
2.771	24.9948	2387	2.7175	0.0165	0.9970	0.8258
2.7049	26.0	2483	2.6822	0.0165	1.0260	0.7733

Framework versions

• Transformers 4.44.2
• Pytorch 2.1.0+cu118
• Datasets 2.20.0
• Tokenizers 0.19.1