wav2vec2-base-checkpoint-7.1

This model is a fine-tuned version of jiobiala24/wav2vec2-base-checkpoint-6 on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.9369
• Wer: 0.3243

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
0.3124	1.75	1000	0.5602	0.3403
0.2428	3.5	2000	0.5924	0.3431
0.1884	5.24	3000	0.6161	0.3423
0.1557	6.99	4000	0.6570	0.3415
0.1298	8.74	5000	0.6837	0.3446
0.1141	10.49	6000	0.7304	0.3396
0.1031	12.24	7000	0.7264	0.3410
0.0916	13.99	8000	0.7229	0.3387
0.0835	15.73	9000	0.8078	0.3458
0.0761	17.48	10000	0.8304	0.3408
0.0693	19.23	11000	0.8290	0.3387
0.0646	20.98	12000	0.8593	0.3372
0.0605	22.73	13000	0.8728	0.3345
0.0576	24.48	14000	0.9111	0.3297
0.0529	26.22	15000	0.9247	0.3273
0.0492	27.97	16000	0.9248	0.3250
0.0472	29.72	17000	0.9369	0.3243

Framework versions

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