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

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: 0.4583
• Wer: 0.3999

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: 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
2.8991	0.55	100	2.9364	1.0000
2.7986	1.09	200	2.8196	0.9999
1.8336	1.64	300	1.1436	0.7846
1.1463	2.19	400	0.7366	0.4903
0.9671	2.73	500	0.6065	0.4828
0.8179	3.28	600	0.6345	0.4917
0.6773	3.83	700	0.5766	0.4867
0.8431	4.37	800	0.6140	0.4933
0.6788	4.92	900	0.5155	0.4624
0.6595	5.46	1000	0.5325	0.4545
0.8755	6.01	1100	0.5265	0.4610
0.4818	6.56	1200	0.5042	0.4385
0.4767	7.1	1300	0.5227	0.4374
0.9617	7.65	1400	0.5348	0.4347
0.5603	8.2	1500	0.5012	0.4332
0.955	8.74	1600	0.5112	0.4290
0.5913	9.29	1700	0.4789	0.4211
0.7202	9.84	1800	0.4987	0.4197
0.5045	10.38	1900	0.4995	0.4252
0.4371	10.93	2000	0.4892	0.4171
0.3178	11.48	2100	0.4852	0.4114
0.3184	12.02	2200	0.4781	0.4156
0.2477	12.57	2300	0.4767	0.4110
0.3052	13.11	2400	0.4650	0.4082
0.3833	13.66	2500	0.4639	0.4043
0.3854	14.21	2600	0.4725	0.4075
0.2411	14.75	2700	0.4773	0.4040
0.3219	15.3	2800	0.4619	0.4021
0.241	15.85	2900	0.4490	0.4022
0.2043	16.39	3000	0.4624	0.3998
0.211	16.94	3100	0.4669	0.4036
0.1777	17.49	3200	0.4657	0.4010
0.1909	18.03	3300	0.4552	0.4003
0.2134	18.58	3400	0.4584	0.4004
0.1713	19.13	3500	0.4578	0.4000
0.1634	19.67	3600	0.4583	0.3999

Framework versions

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