W2V2-Bert_DigitalUmuganda_Afrivoice_Shona_20hr_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: 0.3305
• Wer: 0.2899
• Cer: 0.0592

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_ratio: 0.1
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.1612	1.0	438	0.3803	0.4293	0.0758
0.2723	2.0	876	0.2631	0.2778	0.0497
0.21	3.0	1314	0.2434	0.2803	0.0495
0.1875	4.0	1752	0.2326	0.2569	0.0458
0.1733	5.0	2190	0.2366	0.2586	0.0448
0.1611	6.0	2628	0.2510	0.2768	0.0478
0.1545	7.0	3066	0.2434	0.2652	0.0461
0.1498	8.0	3504	0.2507	0.2947	0.0485
0.1421	9.0	3942	0.2451	0.2596	0.0460
0.1359	10.0	4380	0.2422	0.2603	0.0471
0.1279	11.0	4818	0.2527	0.2511	0.0449
0.116	12.0	5256	0.2667	0.2547	0.0451
0.1044	13.0	5694	0.2630	0.2891	0.0477
0.0927	14.0	6132	0.2811	0.2600	0.0448
0.0843	15.0	6570	0.2852	0.2666	0.0461
0.0725	16.0	7008	0.2936	0.2651	0.0456
0.0639	17.0	7446	0.3091	0.2682	0.0468
0.0552	18.0	7884	0.3164	0.2593	0.0467
0.0473	19.0	8322	0.3319	0.2684	0.0466
0.0429	20.0	8760	0.3389	0.2734	0.0474
0.0366	21.0	9198	0.3646	0.2777	0.0468

Framework versions

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