wav2vec2-large-lv60-200-BPE

This model is a fine-tuned version of facebook/wav2vec2-large-lv60 on the None dataset. It achieves the following results on the evaluation set:

• Loss: nan
• Wer: 1.0

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: 0.0005
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1500
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.5404	0.69	1000	4.7838	1.0
1.4467	1.39	2000	1.4455	0.9006
1.0879	2.08	3000	1.1046	0.6413
0.9838	2.77	4000	1.0398	0.6339
0.8974	3.47	5000	1.0120	0.5716
0.8292	4.16	6000	1.0117	0.6164
0.8741	4.85	7000	0.9665	0.5490
0.8567	5.55	8000	1.0425	0.5486
0.0	6.24	9000	nan	1.0
0.0	6.93	10000	nan	1.0
0.0	7.63	11000	nan	1.0
0.0	8.32	12000	nan	1.0
0.0	9.02	13000	nan	1.0
0.0	9.71	14000	nan	1.0
0.0	10.4	15000	nan	1.0
0.0	11.1	16000	nan	1.0
0.0	11.79	17000	nan	1.0
0.0	12.48	18000	nan	1.0
0.0	13.18	19000	nan	1.0
0.0	13.87	20000	nan	1.0
0.0	14.56	21000	nan	1.0
0.0	15.26	22000	nan	1.0
0.0	15.95	23000	nan	1.0
0.0	16.64	24000	nan	1.0
0.0	17.34	25000	nan	1.0
0.0	18.03	26000	nan	1.0
0.0	18.72	27000	nan	1.0
0.0	19.42	28000	nan	1.0

Framework versions

• Transformers 4.26.1
• Pytorch 1.13.1+cu117
• Datasets 2.10.1
• Tokenizers 0.13.2