w2v-bert-2.0_BIG-C_corpus_Bemba_1hr_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: 0.5722
• Wer: 0.5985
• Cer: 0.1196

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: 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
7.0053	1.0	15	6.3978	1.0085	1.1686
5.7304	2.0	30	4.1429	1.0	0.9477
3.6107	3.0	45	3.1577	1.0	0.9999
3.087	4.0	60	2.8811	1.0	0.9799
2.8587	5.0	75	2.7459	1.0	0.9090
2.7163	6.0	90	2.4646	1.0	0.8254
2.3267	7.0	105	1.9810	0.9998	0.6766
1.7199	8.0	120	1.3007	0.9820	0.3001
1.1524	9.0	135	0.9403	0.8640	0.2081
0.9037	10.0	150	0.8880	0.7630	0.1692
0.7707	11.0	165	0.7744	0.7416	0.1788
0.6817	12.0	180	0.7403	0.6391	0.1462
0.6124	13.0	195	0.7595	0.6170	0.1406
0.5606	14.0	210	0.7323	0.6665	0.1565
0.5283	15.0	225	0.7329	0.7097	0.1781
0.4703	16.0	240	0.7322	0.6011	0.1402
0.5413	17.0	255	0.7942	0.7116	0.1545
0.531	18.0	270	0.8518	0.6595	0.1535
0.5132	19.0	285	0.8821	0.6633	0.1442
0.4961	20.0	300	0.7836	0.6450	0.1478
0.5584	21.0	315	0.9809	0.6544	0.1546
0.7199	22.0	330	0.9238	0.7732	0.2111
0.8428	23.0	345	0.8865	0.7223	0.1874
0.9216	24.0	360	1.3912	0.9975	0.6157
1.1638	25.0	375	1.1943	0.7590	0.1808
1.0508	26.0	390	1.1233	0.9919	0.4404

Framework versions

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