wav2vec2-large-mms-1b-even-pakendorf

This model is a fine-tuned version of facebook/mms-1b-all on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: inf
• Wer: 0.7591
• Cer: 0.2779

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: 0.001
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.847	0.1895	300	inf	0.9027	0.3662
1.8253	0.3790	600	inf	0.9087	0.3658
1.6956	0.5685	900	inf	0.8723	0.3412
1.6616	0.7581	1200	inf	0.8437	0.3209
1.5962	0.9476	1500	inf	0.8392	0.3217
1.6299	1.1371	1800	inf	0.8447	0.3201
1.5242	1.3266	2100	inf	0.8191	0.3076
1.582	1.5161	2400	inf	0.8157	0.3070
1.5555	1.7056	2700	inf	0.8092	0.3061
1.5476	1.8951	3000	inf	0.7999	0.3009
1.4725	2.0846	3300	inf	0.7945	0.2952
1.4902	2.2742	3600	inf	0.7834	0.2936
1.3984	2.4637	3900	inf	0.7836	0.2900
1.4633	2.6532	4200	inf	0.7942	0.2872
1.4533	2.8427	4500	inf	0.7804	0.2863
1.4814	3.0322	4800	inf	0.7728	0.2859
1.4397	3.2217	5100	inf	0.7693	0.2818
1.4218	3.4112	5400	inf	0.7702	0.2831
1.3655	3.6008	5700	inf	0.7650	0.2795
1.34	3.7903	6000	inf	0.7615	0.2792
1.3351	3.9798	6300	inf	0.7591	0.2779

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.1+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1