wav2vec2-common_voice-tr-demo-dist

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the COMMON_VOICE - TR dataset. It achieves the following results on the evaluation set:

• Loss: 0.3848
• Wer: 0.3242

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: 0.0003
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 16
• total_eval_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: 15.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.5279	0.46	100	3.6260	1.0
3.1065	0.92	200	3.0854	0.9999
1.4111	1.38	300	1.3343	0.8839
0.8468	1.83	400	0.6920	0.6826
0.6242	2.29	500	0.6001	0.5996
0.4181	2.75	600	0.5655	0.5680
0.4311	3.21	700	0.4478	0.5003
0.3601	3.67	800	0.4548	0.5011
0.2756	4.13	900	0.4444	0.4682
0.2373	4.59	1000	0.4111	0.4432
0.1831	5.05	1100	0.4178	0.4447
0.2423	5.5	1200	0.3881	0.4277
0.2128	5.96	1300	0.3865	0.4018
0.1256	6.42	1400	0.3818	0.4137
0.1038	6.88	1500	0.3739	0.3942
0.1662	7.34	1600	0.3938	0.3929
0.198	7.8	1700	0.3831	0.3837
0.0728	8.26	1800	0.3910	0.3867
0.123	8.72	1900	0.3722	0.3735
0.0776	9.17	2000	0.3938	0.3725
0.1597	9.63	2100	0.3786	0.3697
0.1124	10.09	2200	0.3947	0.3590
0.0965	10.55	2300	0.3952	0.3562
0.0612	11.01	2400	0.3810	0.3476
0.0764	11.47	2500	0.3734	0.3507
0.0973	11.93	2600	0.3935	0.3472
0.0649	12.39	2700	0.3672	0.3413
0.0542	12.84	2800	0.3732	0.3369
0.087	13.3	2900	0.3833	0.3458
0.0196	13.76	3000	0.3761	0.3303
0.0548	14.22	3100	0.3855	0.3274
0.0577	14.68	3200	0.3893	0.3238

Framework versions

• Transformers 4.20.0.dev0
• Pytorch 1.11.0+cu102
• Datasets 2.2.1
• Tokenizers 0.12.1