wav2vec2-xls-r-1b-ka

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the /WORKSPACE/DATA/KA/NOIZY_STUDENT_2/ - KA dataset. It achieves the following results on the evaluation set:

• Loss: 0.1022
• Wer: 0.1527
• Cer: 0.0221

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: 7e-05
• train_batch_size: 16
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.2839	6.45	400	0.2229	0.3609	0.0557
0.9775	12.9	800	0.1271	0.2202	0.0317
0.9045	19.35	1200	0.1268	0.2030	0.0294
0.8652	25.8	1600	0.1211	0.1940	0.0287
0.8505	32.26	2000	0.1192	0.1912	0.0276
0.8168	38.7	2400	0.1086	0.1763	0.0260
0.7737	45.16	2800	0.1098	0.1753	0.0256
0.744	51.61	3200	0.1054	0.1646	0.0239
0.7114	58.06	3600	0.1034	0.1573	0.0228
0.6773	64.51	4000	0.1022	0.1527	0.0221

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.2
• Datasets 1.18.4.dev0
• Tokenizers 0.11.0