yujiepan/internal.wav2vec2-base-superb-ks-int8-structured83

w2v2-ks-jpqd-lr1e-4

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: 0.1228
• Accuracy: 0.9695

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: 64
• seed: 0
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 15.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
2.5357	1.0	399	2.7821	0.6209
2.7107	2.0	798	2.7331	0.6209
2.671	3.0	1197	2.7330	0.6209
14.208	4.0	1596	14.2660	0.7139
21.0916	5.0	1995	21.0315	0.8104
24.4471	6.0	2394	24.2357	0.9073
25.366	7.0	2793	25.0893	0.9273
25.1369	8.0	3192	24.8976	0.9394
0.4678	9.0	3591	0.2528	0.9435
0.3576	10.0	3990	0.1873	0.9613
0.3622	11.0	4389	0.1583	0.9645
0.2796	12.0	4788	0.1419	0.9666
0.3157	13.0	5187	0.1327	0.9693
0.2997	14.0	5586	0.1263	0.9694
0.2667	15.0	5985	0.1228	0.9695

Framework versions

• Transformers 4.26.0
• Pytorch 1.13.1+cu116
• Datasets 2.8.0
• Tokenizers 0.13.2