w2v-bert-2.0-nonstudio_and_studioRecords_final

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: 0.1772
• Wer: 0.1266

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
1.055	0.4601	600	0.3683	0.4608
0.1734	0.9202	1200	0.2620	0.3546
0.1242	1.3804	1800	0.2115	0.3018
0.1075	1.8405	2400	0.2004	0.2889
0.0888	2.3006	3000	0.1870	0.2573
0.078	2.7607	3600	0.1724	0.2267
0.0664	3.2209	4200	0.1572	0.2244
0.0576	3.6810	4800	0.1746	0.2217
0.0522	4.1411	5400	0.1643	0.1796
0.0415	4.6012	6000	0.1781	0.1851
0.0398	5.0613	6600	0.1670	0.1714
0.0301	5.5215	7200	0.1531	0.1617
0.0296	5.9816	7800	0.1463	0.1590
0.0211	6.4417	8400	0.1566	0.1473
0.0206	6.9018	9000	0.1423	0.1468
0.0147	7.3620	9600	0.1443	0.1413
0.0136	7.8221	10200	0.1539	0.1418
0.0105	8.2822	10800	0.1611	0.1383
0.0079	8.7423	11400	0.1761	0.1351
0.0063	9.2025	12000	0.1814	0.1304
0.0043	9.6626	12600	0.1772	0.1266

Framework versions

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