wav2vec2-xls-r-1b-scandinavian-E6-25h-30-epochs-20250208_v6

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: inf
• Wer: 30.3141
• Cer: 10.6693

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 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
1.1557	0.8052	1000	inf	82.2931	29.3959
1.2064	1.6103	2000	inf	60.0400	20.7421
1.0098	2.4155	3000	inf	51.4387	17.2982
0.5777	3.2206	4000	inf	50.0820	16.9330
0.4035	4.0258	5000	inf	49.7392	16.9217
0.4148	4.8309	6000	inf	48.6947	16.5653
0.3909	5.6361	7000	inf	47.2167	16.1521
0.3957	6.4412	8000	inf	46.7938	15.9950
0.5195	7.2464	9000	inf	45.6720	15.6448
0.4977	8.0515	10000	inf	41.3393	14.0356
0.4793	8.8567	11000	inf	41.2713	14.0323
0.4175	9.6618	12000	inf	40.5189	13.7936
0.2602	10.4670	13000	inf	38.5700	13.2449
0.2566	11.2721	14000	inf	37.8977	13.0691
0.2215	12.0773	15000	inf	39.2877	13.5549
0.2557	12.8824	16000	inf	37.8510	12.8671
0.3672	13.6876	17000	inf	38.6660	13.1300
0.3282	14.4928	18000	inf	37.3481	12.7591
0.3101	15.2979	19000	inf	35.9034	12.3669
0.1821	16.1031	20000	inf	35.0097	12.1182
0.1719	16.9082	21000	inf	35.7754	12.4047
0.1529	17.7134	22000	inf	34.9030	12.0927
0.1338	18.5185	23000	inf	34.3040	11.8955
0.2186	19.3237	24000	inf	33.5490	11.7366
0.2311	20.1288	25000	inf	33.3462	11.5700
0.2306	20.9340	26000	inf	32.6766	11.3913
0.2267	21.7391	27000	inf	33.3049	11.4791
0.1381	22.5443	28000	inf	32.1603	11.2660
0.0838	23.3494	29000	inf	32.0910	11.2171
0.0821	24.1546	30000	inf	31.2506	10.9537
0.077	24.9597	31000	inf	31.0865	10.9077
0.1248	25.7649	32000	inf	30.6210	10.7582
0.1548	26.5700	33000	inf	30.7423	10.8239
0.1648	27.3752	34000	inf	30.2901	10.6713
0.1303	28.1804	35000	inf	30.3675	10.6884
0.135	28.9855	36000	inf	30.3075	10.6664
0.062	29.7907	37000	inf	30.3141	10.6693

Framework versions

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