w2v-bert-2.0-Fleurs_AMMI_AFRIVOICE_LRSC-ln-109hrs-v3

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6697
• Wer: 0.1822
• Cer: 0.0578

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: 1e-05
• train_batch_size: 4
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 80
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.2385	1.0	2880	0.3327	0.2364	0.0722
0.4904	2.0	5760	0.2875	0.2059	0.0640
0.4433	3.0	8640	0.2764	0.1968	0.0611
0.4106	4.0	11520	0.2654	0.1877	0.0594
0.3825	5.0	14400	0.2653	0.1925	0.0623
0.3516	6.0	17280	0.2718	0.1814	0.0580
0.3153	7.0	20160	0.2793	0.1844	0.0635
0.2863	8.0	23040	0.2963	0.1804	0.0570
0.263	9.0	25920	0.2943	0.1898	0.0656
0.2431	10.0	28800	0.3205	0.1806	0.0609
0.2249	11.0	31680	0.3262	0.1768	0.0572
0.2113	12.0	34560	0.3392	0.1889	0.0600
0.1963	13.0	37440	0.3715	0.1826	0.0594
0.1833	14.0	40320	0.4002	0.1830	0.0585
0.1696	15.0	43200	0.4091	0.1766	0.0570
0.1576	16.0	46080	0.4199	0.1780	0.0571
0.1456	17.0	48960	0.4295	0.1840	0.0595
0.1334	18.0	51840	0.4817	0.1762	0.0557
0.1216	19.0	54720	0.4826	0.1892	0.0611
0.1105	20.0	57600	0.5200	0.1801	0.0570
0.1	21.0	60480	0.5401	0.1894	0.0592
0.0903	22.0	63360	0.5619	0.1779	0.0556
0.0811	23.0	66240	0.5689	0.1809	0.0574
0.0729	24.0	69120	0.6234	0.1812	0.0570
0.0652	25.0	72000	0.6594	0.1776	0.0565
0.0575	26.0	74880	0.6697	0.1822	0.0578

Framework versions

• Transformers 4.46.3
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.3