xls-r-1b-bemba-genbed-combined

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5103
• Wer: 0.7252

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.1375	100	3.2454	1.0
No log	0.2749	200	2.9592	1.0
No log	0.4124	300	1.0579	1.0633
No log	0.5498	400	0.8291	0.9390
2.3726	0.6873	500	0.9104	0.9572
2.3726	0.8247	600	0.7652	0.8985
2.3726	0.9622	700	0.6910	0.8788
2.3726	1.0997	800	0.5304	0.7518
2.3726	1.2371	900	0.5304	0.7565
0.7002	1.3746	1000	0.5120	0.7657
0.7002	1.5120	1100	0.4937	0.7343
0.7002	1.6495	1200	0.4805	0.7139
0.7002	1.7869	1300	0.4803	0.7177
0.7002	1.9244	1400	0.4576	0.6875
0.5954	2.0619	1500	0.4863	0.7114
0.5954	2.1993	1600	0.5060	0.7316
0.5954	2.3368	1700	0.5103	0.7252

Framework versions

• Transformers 4.46.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0