wav2vec2_xls_r_300m_FLEURS_Shona_1hr_v1

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

• Loss: 12.4333
• Wer: 1.0
• Cer: 0.9978

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	1.0	8	14.4606	2.7557	0.9555
No log	2.0	16	14.3682	1.4924	0.9242
No log	3.0	24	14.1641	1.0992	0.9862
No log	4.0	32	13.8827	1.0229	0.9902
No log	5.0	40	13.4780	1.0076	0.9925
No log	6.0	48	12.8015	1.0	0.9976
No log	7.0	56	11.7499	1.0	1.0
No log	8.0	64	10.0606	1.0	1.0
No log	9.0	72	8.2175	1.0	1.0
No log	10.0	80	6.8124	1.0	1.0
No log	11.0	88	5.9146	1.0	1.0
No log	12.0	96	5.2321	1.0	1.0
10.8148	13.0	104	4.7869	1.0	1.0
10.8148	14.0	112	4.4701	1.0	1.0
10.8148	15.0	120	4.2285	1.0	1.0
10.8148	16.0	128	4.0551	1.0	1.0

Framework versions

• Transformers 4.42.3
• Pytorch 2.1.0+cu118
• Datasets 2.20.0
• Tokenizers 0.19.1