b24-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: 0.3401
• Wer: 0.2625

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: 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
9.4471	0.76	100	3.3151	1.0
3.0392	1.52	200	3.0118	1.0
2.9633	2.29	300	3.0023	1.0
2.9643	3.05	400	2.9365	1.0
2.9381	3.81	500	2.9319	1.0
2.9411	4.58	600	2.9264	1.0
2.9407	5.34	700	2.9141	1.0
2.9027	6.11	800	2.8848	1.0
2.8833	6.87	900	2.8796	0.9988
2.8805	7.63	1000	2.8679	0.9956
2.7051	8.4	1100	1.8944	1.0
1.343	9.16	1200	0.7785	0.6970
0.8156	9.92	1300	0.5659	0.5824
0.591	10.68	1400	0.4982	0.5163
0.488	11.45	1500	0.4421	0.4299
0.4056	12.21	1600	0.3927	0.3959
0.3488	12.97	1700	0.4095	0.3910
0.2977	13.74	1800	0.3833	0.3687
0.273	14.5	1900	0.3690	0.3388
0.2601	15.27	2000	0.3505	0.3121
0.2258	16.03	2100	0.3577	0.3121
0.2122	16.79	2200	0.3467	0.3018
0.2095	17.56	2300	0.3361	0.2951
0.1719	18.32	2400	0.3572	0.2948
0.1722	19.08	2500	0.3380	0.2857
0.1634	19.84	2600	0.3516	0.2883
0.1592	20.61	2700	0.3374	0.2846
0.153	21.37	2800	0.3395	0.2783
0.1479	22.14	2900	0.3336	0.2729
0.1443	22.9	3000	0.3234	0.2669
0.1339	23.66	3100	0.3345	0.2664
0.1149	24.43	3200	0.3369	0.2664
0.1205	25.19	3300	0.3470	0.2660
0.1251	25.95	3400	0.3319	0.2629
0.1201	26.71	3500	0.3381	0.2667
0.1107	27.48	3600	0.3538	0.2655
0.1117	28.24	3700	0.3423	0.2625
0.1104	29.01	3800	0.3398	0.2608
0.104	29.77	3900	0.3401	0.2625

Framework versions

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