W2V2-Bert_DigitalUmuganda_Afrivoice_Shona_50hr_v2

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: 0.2946
• Wer: 0.2791
• Cer: 0.0560

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: 0.03
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.0699	1.0	1092	0.2310	0.2610	0.0441
0.2086	2.0	2184	0.2126	0.2397	0.0416
0.1918	3.0	3276	0.2344	0.2636	0.0468
0.1783	4.0	4368	0.2030	0.2516	0.0420
0.1636	5.0	5460	0.1969	0.2334	0.0402
0.1528	6.0	6552	0.2050	0.2247	0.0393
0.1443	7.0	7644	0.1934	0.2304	0.0389
0.136	8.0	8736	0.1908	0.2292	0.0396
0.1283	9.0	9828	0.1963	0.2255	0.0388
0.1198	10.0	10920	0.1935	0.2216	0.0380
0.1118	11.0	12012	0.2008	0.2229	0.0379
0.1021	12.0	13104	0.1982	0.2233	0.0392
0.0969	13.0	14196	0.2036	0.2245	0.0389
0.0895	14.0	15288	0.2046	0.2238	0.0389
0.0821	15.0	16380	0.2203	0.2225	0.0386
0.0752	16.0	17472	0.2293	0.2221	0.0387
0.0677	17.0	18564	0.2423	0.2430	0.0419
0.0618	18.0	19656	0.2469	0.2262	0.0386
0.0545	19.0	20748	0.2677	0.2372	0.0398
0.0495	20.0	21840	0.2691	0.2295	0.0393

Framework versions

• Transformers 4.44.2
• Pytorch 2.2.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1