b22-wav2vec2-large-xls-r-romansh-colab

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7362
• Wer: 0.4991

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: 0.0003
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.8045	0.76	100	2.9604	1.0
2.9578	1.52	200	3.0626	1.0
2.9565	2.29	300	3.0432	1.0
2.9533	3.05	400	2.9304	1.0
2.9263	3.81	500	2.9134	1.0
2.9174	4.58	600	2.9022	1.0
2.9282	5.34	700	2.8967	1.0
2.8973	6.11	800	2.8477	1.0
2.6047	6.87	900	2.0269	1.0
1.6468	7.63	1000	1.0780	0.9029
1.1006	8.4	1100	0.8305	0.8319
0.8708	9.16	1200	0.7704	0.8055
0.7708	9.92	1300	0.6815	0.7385
0.6608	10.68	1400	0.6738	0.7212
0.6014	11.45	1500	0.6535	0.6940
0.5419	12.21	1600	0.6608	0.6639
0.4961	12.97	1700	0.6568	0.6372
0.4462	13.74	1800	0.6557	0.6362
0.4169	14.5	1900	0.6487	0.5985
0.3951	15.27	2000	0.7126	0.6376
0.3643	16.03	2100	0.6539	0.5859
0.3243	16.79	2200	0.6803	0.5946
0.3243	17.56	2300	0.6619	0.5745
0.2869	18.32	2400	0.6826	0.5592
0.2895	19.08	2500	0.6980	0.5524
0.2612	19.84	2600	0.6599	0.5445
0.2492	20.61	2700	0.6533	0.5394
0.2485	21.37	2800	0.7103	0.5494
0.2352	22.14	2900	0.7339	0.5501
0.2136	22.9	3000	0.7154	0.5470
0.2079	23.66	3100	0.7360	0.5389
0.2011	24.43	3200	0.7481	0.5263
0.1925	25.19	3300	0.7409	0.5186
0.193	25.95	3400	0.7334	0.5091
0.1874	26.71	3500	0.7493	0.5075
0.1802	27.48	3600	0.7362	0.5102
0.1736	28.24	3700	0.7427	0.5033
0.1725	29.01	3800	0.7404	0.5033
0.1684	29.77	3900	0.7362	0.4991

Framework versions

• Transformers 4.26.0
• Pytorch 2.0.1+cu118
• Datasets 2.14.4
• Tokenizers 0.13.3