wav2vec2-xls-r-1b-dutch-2

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: 0.4279
• Wer: 0.3959

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: 9e-05
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.02
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.8915	0.55	100	2.9772	0.9996
2.8043	1.09	200	2.7246	0.9999
1.5968	1.64	300	0.8340	0.6015
1.087	2.19	400	0.6961	0.4067
0.9196	2.73	500	0.5938	0.4736
0.7747	3.28	600	0.5167	0.4576
0.6913	3.83	700	0.5408	0.4511
0.8202	4.37	800	0.5513	0.4687
0.6635	4.92	900	0.4636	0.4533
0.6634	5.46	1000	0.4891	0.4541
0.8651	6.01	1100	0.4832	0.4275
0.4688	6.56	1200	0.4799	0.4297
0.4639	7.1	1300	0.4912	0.4358
0.9784	7.65	1400	0.5018	0.4228
0.5491	8.2	1500	0.4799	0.4335
0.9546	8.74	1600	0.4587	0.4333
0.799	9.29	1700	0.4456	0.4226
0.9198	9.84	1800	0.4499	0.4201
0.6325	10.38	1900	0.4916	0.4301
0.5124	10.93	2000	0.4728	0.4198
0.354	11.48	2100	0.4640	0.4188
0.3738	12.02	2200	0.4518	0.4133
0.2682	12.57	2300	0.4406	0.4094
0.3651	13.11	2400	0.4382	0.4074
0.3935	13.66	2500	0.4373	0.4038
0.4126	14.21	2600	0.4354	0.4036
0.2614	14.75	2700	0.4363	0.3979
0.3284	15.3	2800	0.4317	0.3962
0.2693	15.85	2900	0.4206	0.3977
0.2211	16.39	3000	0.4244	0.3974
0.2234	16.94	3100	0.4311	0.3980
0.2001	17.49	3200	0.4253	0.3957
0.2005	18.03	3300	0.4257	0.3962
0.2214	18.58	3400	0.4278	0.3961
0.1843	19.13	3500	0.4277	0.3959
0.1692	19.67	3600	0.4279	0.3959

Framework versions

• Transformers 4.39.3
• Pytorch 2.7.0+cu126
• Datasets 2.16.1
• Tokenizers 0.15.2