XLS-R-SWAHILI-ASR-CV14

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_14_0 dataset. It achieves the following results on the evaluation set:

• Loss: inf
• Wer: 0.2148
• Cer: 0.0684

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: 0.0003
• 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
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
3.9008	0.33	400	0.2565	inf	0.8327
0.5689	0.66	800	0.1306	inf	0.4598
0.3838	1.0	1200	0.1130	inf	0.3786
0.3054	1.33	1600	0.1032	inf	0.3407
0.2877	1.66	2000	0.0976	inf	0.3239
0.2698	1.99	2400	0.0952	inf	0.3078
0.2285	2.32	2800	0.0956	inf	0.3031
0.224	2.66	3200	0.0892	inf	0.2861
0.2224	2.99	3600	0.0877	inf	0.2809
0.1906	3.32	4000	0.0853	inf	0.2748
0.1897	3.65	4400	0.0844	inf	0.2707
0.183	3.98	4800	0.0814	inf	0.2614
0.1586	4.32	5200	0.0809	inf	0.2569
0.162	4.65	5600	0.0782	inf	0.2493
0.1548	4.98	6000	0.0772	inf	0.2467
0.1364	5.31	6400	0.0782	inf	0.2459
0.1344	5.64	6800	0.0760	inf	0.2404
0.1301	5.98	7200	0.0738	inf	0.2346
0.1165	6.31	7600	inf	0.2321	0.0729
0.1142	6.64	8000	inf	0.2266	0.0719
0.1103	6.97	8400	inf	0.2229	0.0705
0.101	7.3	8800	inf	0.2203	0.0699
0.1006	7.63	9200	inf	0.2174	0.0692
0.0958	7.97	9600	inf	0.2160	0.0688
0.0896	8.3	10000	inf	0.2148	0.0684

Framework versions

• Transformers 4.38.1
• Pytorch 2.2.1
• Datasets 2.17.0
• Tokenizers 0.15.2