wav2vec2-xls-r-300m-scandinavian-E4-100h-30-epochs-20250201_v2.2

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5180
• Wer: 33.7460
• Cer: 9.2540

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: 5e-05
• 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: 5000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.3709	0.9153	1000	3.2919	100.0	99.9998
2.9006	1.8302	2000	2.8985	100.0	99.9998
1.0856	2.7451	3000	0.7243	59.2040	17.1565
0.6157	3.6600	4000	0.3552	34.2313	9.4613
0.4596	4.5748	5000	0.2829	27.9991	7.6917
0.4002	5.4897	6000	0.2588	26.1989	7.0489
0.8133	6.4046	7000	0.5522	30.9377	8.5844
1.244	7.3195	8000	1.0083	71.3211	25.5473
0.9065	8.2343	9000	0.6510	35.2864	9.6846
0.7912	9.1492	10000	0.5297	33.3216	9.1039
0.7898	10.0641	11000	0.5132	33.3569	9.1523
0.7686	10.9794	12000	0.5126	33.5706	9.1892
0.7419	11.8943	13000	0.5180	33.7460	9.2549
0.7229	12.8092	14000	0.5180	33.7438	9.2557
0.7368	13.7240	15000	0.5180	33.7481	9.2555
0.7861	14.6389	16000	0.5180	33.7385	9.2536
0.7975	15.5538	17000	0.5180	33.7406	9.2549
0.8011	16.4686	18000	0.5180	33.7481	9.2546
0.7773	17.3835	19000	0.5180	33.7460	9.2531
0.7719	18.2984	20000	0.5180	33.7449	9.2546
0.7456	19.2133	21000	0.5180	33.7449	9.2542
0.7024	20.1281	22000	0.5180	33.7449	9.2535
0.7654	21.0430	23000	0.5180	33.7449	9.2546
0.8013	21.9584	24000	0.5180	33.7460	9.2570
0.823	22.8732	25000	0.5180	33.7449	9.2551
0.8084	23.7881	26000	0.5180	33.7438	9.2536
0.8533	24.7030	27000	0.5180	33.7417	9.2533
0.7409	25.6178	28000	0.5180	33.7438	9.2535
0.7502	26.5327	29000	0.5180	33.7502	9.2548
0.7424	27.4476	30000	0.5180	33.7492	9.2559
0.7426	28.3625	31000	0.5180	33.7428	9.2538
0.7798	29.2773	32000	0.5180	33.7460	9.2540

Framework versions

• Transformers 4.48.2
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0