wav2vec2-large-xls-r-300m-lg-20hr-v3

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5269
• Wer: 0.6523
• Cer: 0.1443

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	0.9948	95	4.7296	1.0	1.0
10.707	2.0	191	3.2782	1.0	1.0
3.6293	2.9948	286	2.9509	1.0	1.0
3.0251	4.0	382	2.5502	1.0	0.8614
2.7179	4.9948	477	0.8813	0.9115	0.2460
1.0647	6.0	573	0.6062	0.8192	0.1791
0.5597	6.9948	668	0.4977	0.7163	0.1474
0.3926	8.0	764	0.4691	0.6665	0.1396
0.292	8.9948	859	0.4583	0.6417	0.1298
0.2443	9.9476	950	0.4634	0.6453	0.1309

Framework versions

• Transformers 4.42.3
• Pytorch 2.2.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1