wav2vec2-base-ks-linear_lrX1000

This model is a fine-tuned version of facebook/wav2vec2-base on the superb dataset. It achieves the following results on the evaluation set:

• Loss: 0.5661
• Accuracy: 0.8325

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: 0.03
• train_batch_size: 256
• eval_batch_size: 256
• seed: 0
• gradient_accumulation_steps: 4
• total_train_batch_size: 1024
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.7558	1.0	50	1.0584	0.6462
0.5971	2.0	100	0.7816	0.7510
0.5382	3.0	150	0.7870	0.7520
0.5045	4.0	200	0.6647	0.7880
0.4717	5.0	250	1.1572	0.6053
0.4651	6.0	300	0.6387	0.7945
0.4205	7.0	350	0.5661	0.8325
0.4423	8.0	400	0.7100	0.7846
0.426	9.0	450	0.7054	0.7829
0.4067	10.0	500	0.6288	0.8114

Framework versions

• Transformers 4.22.0.dev0
• Pytorch 1.11.0+cu115
• Datasets 2.4.0
• Tokenizers 0.12.1