wav2vec2-base-checkpoint-8

This model is a fine-tuned version of jiobiala24/wav2vec2-base-checkpoint-7.1 on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.9561
• Wer: 0.3271

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: 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
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3117	1.59	1000	0.5514	0.3451
0.2509	3.19	2000	0.5912	0.3328
0.1918	4.78	3000	0.6103	0.3346
0.1612	6.38	4000	0.6469	0.3377
0.1388	7.97	5000	0.6597	0.3391
0.121	9.57	6000	0.6911	0.3472
0.1096	11.16	7000	0.7300	0.3457
0.0959	12.76	8000	0.7660	0.3400
0.0882	14.35	9000	0.8316	0.3394
0.0816	15.95	10000	0.8042	0.3357
0.0739	17.54	11000	0.8087	0.3346
0.0717	19.14	12000	0.8590	0.3353
0.066	20.73	13000	0.8750	0.3336
0.0629	22.33	14000	0.8759	0.3333
0.0568	23.92	15000	0.8963	0.3321
0.0535	25.52	16000	0.9391	0.3323
0.0509	27.11	17000	0.9279	0.3296
0.0498	28.71	18000	0.9561	0.3271

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu111
• Datasets 1.13.3
• Tokenizers 0.10.3