Dzo_ASR_xlsr_53

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: 0.5489
• Wer: 0.1957

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: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.7643	1.7544	400	3.3975	1.0
3.2365	3.5088	800	3.0763	0.9975
3.0169	5.2632	1200	2.3831	1.0191
1.4481	7.0175	1600	0.8834	0.5629
0.8228	8.7719	2000	0.6189	0.3965
0.6329	10.5263	2400	0.5400	0.3333
0.5043	12.2807	2800	0.5054	0.2831
0.4179	14.0351	3200	0.4820	0.2625
0.3599	15.7895	3600	0.4979	0.2406
0.3152	17.5439	4000	0.4816	0.2286
0.2778	19.2982	4400	0.4848	0.2183
0.2557	21.0526	4800	0.4710	0.2172
0.2285	22.8070	5200	0.5168	0.2053
0.2113	24.5614	5600	0.5015	0.2013
0.1992	26.3158	6000	0.5347	0.1989
0.1846	28.0702	6400	0.5403	0.1974
0.1804	29.8246	6800	0.5489	0.1957

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 2.14.4
• Tokenizers 0.21.1