w2v2-base-pretrained_lr5e-5_at0.8_da0.7

This model is a fine-tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.2042
• Wer: 0.1884

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: 32
• 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: 1000
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
24.3422	7.81	250	5.6070	1.0
3.5981	15.62	500	3.2535	1.0
3.1121	23.44	750	3.1577	1.0
3.0596	31.25	1000	3.1214	1.0
3.0143	39.06	1250	2.9603	1.0
1.4861	46.88	1500	1.2406	0.4007
0.2223	54.69	1750	1.3926	0.2324
0.1147	62.5	2000	1.5275	0.2136
0.0775	70.31	2250	1.8277	0.1986
0.0601	78.12	2500	1.9747	0.1944
0.0479	85.94	2750	2.0632	0.1909
0.042	93.75	3000	2.1333	0.1991
0.0353	101.56	3250	2.1743	0.1982
0.0315	109.38	3500	2.1585	0.1939
0.0274	117.19	3750	2.1521	0.1914
0.0279	125.0	4000	2.2042	0.1884

Framework versions

• Transformers 4.35.0
• Pytorch 2.0.0
• Datasets 2.14.6
• Tokenizers 0.14.1