w2v2-lmk_updated_auto

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: 1.3904
• Wer: 0.5923
• Cer: 0.2399

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 OptimizerNames.ADAMW_TORCH_FUSED 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
8.4649	4.7619	100	4.0812	1.0	1.0
3.2496	9.5238	200	2.9563	1.0	1.0
2.9748	14.2857	300	2.9203	1.0	1.0
2.8994	19.0476	400	2.8260	1.0	1.0
2.5637	23.8095	500	2.1752	1.0	0.8309
1.8029	28.5714	600	1.4903	0.9861	0.5156
1.4084	33.3333	700	1.2974	0.9164	0.3869
1.1087	38.0952	800	1.2984	0.7387	0.2856
0.9667	42.8571	900	1.2944	0.6794	0.2727
0.8039	47.6190	1000	1.3522	0.6481	0.2681
0.7177	52.3810	1100	1.2763	0.5993	0.2353
0.6455	57.1429	1200	1.2724	0.6446	0.2475
0.5526	61.9048	1300	1.3468	0.6411	0.2483
0.5185	66.6667	1400	1.3384	0.6167	0.2414
0.498	71.4286	1500	1.3211	0.6098	0.2475
0.4556	76.1905	1600	1.3733	0.6272	0.2430
0.4398	80.9524	1700	1.4148	0.6376	0.2559
0.416	85.7143	1800	1.3720	0.6132	0.2437
0.4177	90.4762	1900	1.3502	0.5923	0.2407
0.4069	95.2381	2000	1.3766	0.6028	0.2460
0.4045	100.0	2100	1.3904	0.5923	0.2399

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 3.0.0
• Tokenizers 0.22.1