wav2vec2-xls-r-1b-irish-5h-11k-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: nan
• Wer: 99.9711
• Cer: 99.9943

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
2.9708	6.25	1000	2.8622	94.9986	93.1290
4.3506	12.5	2000	4.2474	96.5019	92.1035
4.4125	18.75	3000	4.2473	96.5019	92.2060
4.4092	25.0	4000	4.2474	96.3862	91.9781
4.436	31.25	5000	4.2474	96.4151	92.0978
4.4694	37.5	6000	4.2473	96.4441	92.1206
4.3806	43.75	7000	4.2474	96.3862	92.1775
4.4963	50.0	8000	4.2473	96.3862	92.0579
0.0	56.25	9000	nan	99.9711	99.9943
0.0	62.5	10000	nan	99.9711	99.9943
0.0	68.75	11000	nan	99.9711	99.9943

Framework versions

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