w2v-bert-2.0-swahili-colab-CV16.0_5epochs

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the common_voice_16_0 dataset. It achieves the following results on the evaluation set:

• Loss: inf
• Wer: 0.8219

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: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.015	0.16	300	inf	0.2387
0.2497	0.33	600	inf	0.2413
0.2246	0.49	900	inf	0.2121
0.2032	0.66	1200	inf	0.2097
0.1895	0.82	1500	inf	0.1969
0.1897	0.99	1800	inf	0.2092
0.1718	1.15	2100	inf	0.1895
0.1872	1.31	2400	inf	0.1949
0.2056	1.48	2700	inf	0.1975
0.3533	1.64	3000	inf	0.4304
0.5492	1.81	3300	inf	0.2979
1.0312	1.97	3600	inf	0.5560
0.8936	2.14	3900	inf	0.8217
1.0655	2.3	4200	inf	0.8219
1.0856	2.46	4500	inf	0.8219
1.0855	2.63	4800	inf	0.8219
1.0823	2.79	5100	inf	0.8219
1.0847	2.96	5400	inf	0.8219
1.0835	3.12	5700	inf	0.8219
1.0886	3.28	6000	inf	0.8219
1.0801	3.45	6300	inf	0.8219
1.0765	3.61	6600	inf	0.8219
1.0878	3.78	6900	inf	0.8219
1.0884	3.94	7200	inf	0.8219
1.0824	4.11	7500	inf	0.8219
1.0881	4.27	7800	inf	0.8219
1.0884	4.43	8100	inf	0.8219
1.0786	4.6	8400	inf	0.8219
1.0846	4.76	8700	inf	0.8219
1.0861	4.93	9000	inf	0.8219

Framework versions

• Transformers 4.37.1
• Pytorch 2.1.1+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0