wav2vec2-jasmin-nl-vl-dia-read-xls-r-300m

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

• Loss: 0.4013
• Wer: 1.0930

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: 64
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 3
• total_train_batch_size: 192
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15.0

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.19	100	4.5749	1.0
No log	2.38	200	3.2544	1.0
No log	3.57	300	3.1107	1.0001
No log	4.76	400	0.9152	1.0171
3.9561	5.95	500	0.6422	1.0590
3.9561	7.14	600	0.4900	1.0867
3.9561	8.33	700	0.4251	1.0955
3.9561	9.52	800	0.4232	1.1465
3.9561	10.71	900	0.4133	1.1194
0.2637	11.9	1000	0.3940	1.1307
0.2637	13.1	1100	0.3943	1.0964
0.2637	14.29	1200	0.3974	1.0894

Framework versions

• Transformers 4.32.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.14.1
• Tokenizers 0.13.3