wav2vec2-large-xls-r-300m-amharic-demo-colab

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_16_1 dataset. It achieves the following results on the evaluation set:

• Loss: 2.0166
• Wer: 0.9159

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 80
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
12.6278	5.0	100	4.1583	1.0
4.0971	10.0	200	4.0317	1.0
3.9986	15.0	300	3.9758	1.0
3.7669	20.0	400	3.2290	1.0287
1.6097	25.0	500	1.8216	0.9860
0.5931	30.0	600	1.7982	0.9780
0.3501	35.0	700	1.9234	0.9867
0.2629	40.0	800	1.9051	0.9206
0.2055	45.0	900	1.9681	0.9246
0.1844	50.0	1000	2.0111	0.9393
0.1625	55.0	1100	2.0117	0.9286
0.1486	60.0	1200	2.0144	0.9326
0.1348	65.0	1300	2.0011	0.9373
0.1183	70.0	1400	2.0303	0.9053
0.1095	75.0	1500	2.0183	0.9239
0.1064	80.0	1600	2.0166	0.9159

Framework versions

• Transformers 4.42.0
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1