wav2vec2_xls_r_300m_Afrivoice_FLEURS_Shona_100hr_v1

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

• Loss: 0.2326
• Wer: 0.2659
• Cer: 0.0491

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: 5e-05
• train_batch_size: 4
• eval_batch_size: 4
• 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: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.5924	0.9998	2224	0.2704	0.3730	0.0679
0.2193	2.0	4449	0.2025	0.2838	0.0484
0.1765	2.9998	6673	0.1895	0.2641	0.0444
0.1533	4.0	8898	0.1875	0.2694	0.0449
0.1355	4.9998	11122	0.1857	0.2502	0.0419
0.1188	6.0	13347	0.1924	0.2520	0.0423
0.1048	6.9998	15571	0.1923	0.2509	0.0423
0.0915	8.0	17796	0.2057	0.2545	0.0423
0.0793	8.9998	20020	0.2071	0.2575	0.0428
0.069	10.0	22245	0.2214	0.2571	0.0424
0.0591	10.9998	24469	0.2346	0.2624	0.0422
0.051	12.0	26694	0.2474	0.2614	0.0432
0.0446	12.9998	28918	0.2598	0.2675	0.0429
0.0396	14.0	31143	0.2689	0.2647	0.0429
0.0353	14.9998	33367	0.2745	0.2645	0.0422

Framework versions

• Transformers 4.45.2
• Pytorch 2.1.0+cu118
• Datasets 3.0.1
• Tokenizers 0.20.1