W2V2-Bert_DigitalUmuganda_Afrivoice_Shona_10hr_v1

This model is a fine-tuned version of facebook/w2v-bert-2.0 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3724
• Wer: 0.3126
• Cer: 0.0617

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: 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: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.241	0.9977	220	0.3383	0.3722	0.0674
0.2577	2.0	441	0.3071	0.3285	0.0612
0.2209	2.9977	661	0.2699	0.3233	0.0565
0.1901	4.0	882	0.2798	0.3381	0.0570
0.1665	4.9977	1102	0.2710	0.2862	0.0513
0.1479	6.0	1323	0.2761	0.2970	0.0530
0.1323	6.9977	1543	0.2726	0.2888	0.0508
0.1172	8.0	1764	0.2843	0.2831	0.0505
0.1021	8.9977	1984	0.2944	0.2760	0.0497
0.0901	10.0	2205	0.3071	0.3010	0.0526
0.0794	10.9977	2425	0.3066	0.2766	0.0487
0.0673	12.0	2646	0.3258	0.2884	0.0511
0.0572	12.9977	2866	0.3349	0.2940	0.0517
0.0497	14.0	3087	0.3684	0.2954	0.0510
0.0428	14.9977	3307	0.3969	0.2792	0.0498
0.035	16.0	3528	0.4034	0.3074	0.0518
0.0317	16.9977	3748	0.4106	0.2874	0.0501
0.0284	18.0	3969	0.4286	0.2845	0.0516
0.0227	18.9977	4189	0.4147	0.2959	0.0497

Framework versions

• Transformers 4.44.1
• Pytorch 2.2.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1