asr-nepali

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: nan
• Wer: 1.0
• Cer: 0.9965
• Accuracy: 0.0035

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 180
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Accuracy
451.545	1.46	100	43.3285	1.0	0.9684	0.0316
194.4567	2.92	200	nan	1.0	0.9965	0.0035
0.0	4.38	300	nan	1.0	0.9965	0.0035
0.0	5.84	400	nan	1.0	0.9965	0.0035
0.0	7.3	500	nan	1.0	0.9965	0.0035
0.0	8.76	600	nan	1.0	0.9965	0.0035
0.0	10.22	700	nan	1.0	0.9965	0.0035
0.0	11.68	800	nan	1.0	0.9965	0.0035
0.0	13.14	900	nan	1.0	0.9965	0.0035
0.0	14.6	1000	nan	1.0	0.9965	0.0035
0.0	16.06	1100	nan	1.0	0.9965	0.0035
0.0	17.52	1200	nan	1.0	0.9965	0.0035
0.0	18.98	1300	nan	1.0	0.9965	0.0035
0.0	20.44	1400	nan	1.0	0.9965	0.0035
0.0	21.9	1500	nan	1.0	0.9965	0.0035
0.0	23.36	1600	nan	1.0	0.9965	0.0035
0.0	24.82	1700	nan	1.0	0.9965	0.0035
0.0	26.28	1800	nan	1.0	0.9965	0.0035
0.0	27.74	1900	nan	1.0	0.9965	0.0035
0.0	29.2	2000	nan	1.0	0.9965	0.0035

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0