wav2vec2-xls-r-1b-danish-12h-6k-steps

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4179
• Wer: 29.8051
• Cer: 9.5826

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 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: 3000
• training_steps: 11000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.9182	5.3333	1000	0.5134	53.0768	16.3044
0.2894	10.6667	2000	0.3309	35.3529	10.9777
0.2917	16.0	3000	0.3877	38.0657	12.0348
0.1964	21.3333	4000	0.4244	36.1713	11.4545
0.1227	26.6667	5000	0.4213	36.4335	11.6030
0.1455	32.0	6000	0.4112	34.1412	10.9986
0.1005	37.3333	7000	0.4383	33.8563	10.8228
0.0604	42.6667	8000	0.4381	33.0379	10.5787
0.0616	48.0	9000	0.4445	31.4826	10.0955
0.0425	53.3333	10000	0.4412	30.7637	9.8170
0.0326	58.6667	11000	0.4179	29.8051	9.5826

Framework versions

• Transformers 4.46.2
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3