hubert_large_528_10

This model is a fine-tuned version of facebook/hubert-large-ls960-ft on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.7757
• Wer: 0.2425

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: 0.0001
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 8
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.1878	0.3378	100	0.9689	0.3245
1.2219	0.6757	200	0.8869	0.2887
1.4146	1.0135	300	0.7800	0.2827
0.9472	1.3514	400	0.7899	0.2829
0.9176	1.6892	500	0.8345	0.2773
1.0975	2.0270	600	0.8103	0.2704
1.0267	2.3649	700	0.8043	0.2661
0.9401	2.7027	800	0.8001	0.2648
0.9752	3.0405	900	0.8112	0.2595
0.9301	3.3784	1000	0.8174	0.2593
0.7361	3.7162	1100	0.8497	0.2567
0.9846	4.0541	1200	0.8002	0.2513
0.9202	4.3919	1300	0.7937	0.2524
0.7069	4.7297	1400	0.8582	0.2448
1.0649	5.0676	1500	0.7993	0.2449
0.6096	5.4054	1600	0.8183	0.2442
0.6876	5.7432	1700	0.8041	0.2426
0.7614	6.0811	1800	0.8133	0.2454
0.572	6.4189	1900	0.7747	0.2441
0.5909	6.7568	2000	0.7610	0.2447
0.562	7.0946	2100	0.7784	0.2450
0.5979	7.4324	2200	0.7675	0.2427
0.541	7.7703	2300	0.7757	0.2425

Framework versions

• Transformers 4.41.0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1