wav2vec2-xls-r-1b-scandinavian-E5-100h-30-epochs-20250124

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.2093
• Wer: 31.5873
• Cer: 16.9806

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: 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: cosine
• lr_scheduler_warmup_steps: 6000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.3694	2.6324	1000	1.1761	67.3147	26.5833
0.8825	5.2635	2000	0.8142	38.0987	17.7731
0.771	7.8959	3000	0.7671	36.3361	17.4212
0.6715	10.5270	4000	0.8356	33.9325	17.2121
0.6258	13.1581	5000	0.7917	35.6939	17.7429
0.5753	15.7905	6000	0.8691	35.4330	17.9205
0.4999	18.4216	7000	0.8614	34.5560	17.5324
0.4432	21.0527	8000	0.9074	33.6095	17.4453
0.3479	23.6851	9000	1.0258	33.6368	17.7273
0.3202	26.3162	10000	1.1457	31.8084	17.0006
0.2974	28.9486	11000	1.2093	31.5873	16.9806

Framework versions

• Transformers 4.48.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0