luganda_wav2vec2_ctc_train_clean

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_7_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2835
• Wer: 0.4156

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=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.8861	2.4	500	3.1284	1.0
2.0448	4.81	1000	0.5439	0.7131
0.6342	7.21	1500	0.3713	0.5556
0.4907	9.62	2000	0.3464	0.5015
0.4242	12.02	2500	0.3122	0.4746
0.3898	14.42	3000	0.3164	0.4634
0.357	16.83	3500	0.2896	0.4416
0.3338	19.23	4000	0.2880	0.4409
0.3223	21.63	4500	0.2841	0.4287
0.3072	24.04	5000	0.2849	0.4250
0.2974	26.44	5500	0.2829	0.4194
0.2878	28.85	6000	0.2835	0.4156

Framework versions

• Transformers 4.38.1
• Pytorch 2.2.1+cu121
• Datasets 2.17.0
• Tokenizers 0.15.2