w2v-bert-2.0-wol-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: 0.1008
• Wer: 0.0792

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: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.6351	0.6857	300	0.2974	0.3040
0.4591	1.3714	600	0.2215	0.2307
0.3833	2.0571	900	0.1950	0.1900
0.329	2.7429	1200	0.1637	0.1614
0.2797	3.4286	1500	0.1515	0.1479
0.2558	4.1143	1800	0.1435	0.1337
0.2166	4.8	2100	0.1296	0.1295
0.1876	5.4857	2400	0.1178	0.1129
0.1695	6.1714	2700	0.1107	0.1005
0.137	6.8571	3000	0.1064	0.0933
0.1078	7.5429	3300	0.1049	0.0929
0.0904	8.2286	3600	0.1002	0.0871
0.0685	8.9143	3900	0.0973	0.0810
0.049	9.6	4200	0.1008	0.0792

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.1+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1