w2v2-lmk_org_para

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 1.1860
• Wer: 0.6934
• Cer: 0.2422

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use 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: 300
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
9.5202	6.2581	100	4.3511	1.0	1.0
3.2449	12.5161	200	2.9976	1.0	1.0
3.0171	18.7742	300	2.8768	1.0	1.0
2.9274	25.0	400	2.7959	1.0	1.0
2.6596	31.2581	500	2.3278	1.0	0.9025
1.9635	37.5161	600	1.5400	0.9617	0.4646
1.4694	43.7742	700	1.3102	0.8258	0.3343
1.1617	50.0	800	1.1932	0.8188	0.3092
0.9487	56.2581	900	1.2371	0.7700	0.2833
0.7926	62.5161	1000	1.2255	0.7213	0.2589
0.6905	68.7742	1100	1.1964	0.6864	0.2567
0.6536	75.0	1200	1.1720	0.6864	0.2468
0.5637	81.2581	1300	1.1978	0.6934	0.2483
0.5283	87.5161	1400	1.1845	0.6899	0.2384
0.535	93.7742	1500	1.1884	0.6864	0.2422
0.4905	100.0	1600	1.1860	0.6934	0.2422

Framework versions

• Transformers 4.57.3
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.22.1