w2v-bert-2.0-chichewa_34_68h

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the CLEAR-GLOBAL/CHICHEWA_34_68H - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.2916
• Wer: 0.3797
• Cer: 0.1073

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: 3e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• 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
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 100000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.5711	1.1038	1000	2.8158	0.9906	0.8586
0.1101	2.2076	2000	0.6337	0.5731	0.1691
0.0696	3.3114	3000	0.4766	0.5119	0.1419
0.0329	4.4152	4000	0.4343	0.4797	0.1356
0.0304	5.5191	5000	0.3731	0.4476	0.1293
0.0292	6.6229	6000	0.3978	0.4217	0.1220
0.0644	7.7267	7000	0.3462	0.4160	0.1206
0.2232	8.8305	8000	0.3415	0.4014	0.1167
0.0166	9.9343	9000	0.3268	0.4064	0.1222
0.0216	11.0375	10000	0.3579	0.4225	0.1224
0.0104	12.1414	11000	0.3526	0.4144	0.1208
0.0079	13.2452	12000	0.2968	0.3613	0.1070
0.0116	14.3490	13000	0.3053	0.3904	0.1127
0.0284	15.4528	14000	0.3216	0.3644	0.1069
0.0838	16.5566	15000	0.2916	0.3803	0.1074
0.007	17.6604	16000	0.3104	0.3732	0.1064
0.0039	18.7642	17000	0.3043	0.3715	0.1079
0.0084	19.8680	18000	0.3325	0.3903	0.1114
0.0147	20.9718	19000	0.3620	0.3906	0.1095
0.0083	22.0751	20000	0.3416	0.4119	0.1123

Framework versions

• Transformers 4.48.1
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1