libri-smallw2v2-no-copy-mse-alpha-0.75-T-1-take-5

This model is a fine-tuned version of rohitp1/libri-smallw2v2-no-copy-mse-alpha-0.75-T-1-take-3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 30.3050
• Wer: 0.2650

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: 0.002
• train_batch_size: 4
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.2
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
239.4865	1.12	400	31.0836	0.2908
210.2046	2.25	800	29.7831	0.2742
195.0478	3.37	1200	28.8794	0.2636
188.8096	4.49	1600	28.7458	0.2600
183.6592	5.62	2000	29.1159	0.2573
181.5025	6.74	2400	29.0081	0.2564
181.9954	7.86	2800	28.7132	0.2588
181.8548	8.99	3200	29.3207	0.2630
186.2524	10.11	3600	29.9119	0.2593
188.3838	11.24	4000	30.2963	0.2627
192.2623	12.36	4400	30.3050	0.2650

Framework versions

• Transformers 4.24.0
• Pytorch 1.12.1
• Datasets 2.7.1
• Tokenizers 0.11.0