seq-xls-r-fleurs_nl-run2-asr_af-run2

This model is a fine-tuned version of lucas-meyer/xls-r-fleurs_nl-run2 on the asr_af dataset. It achieves the following results:

• Wer (Validation): 38.75%
• Wer (Test): 38.66%

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 3
• total_train_batch_size: 12
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer (Train)
7.6065	0.44	100	3.3086	1.0
3.055	0.88	200	2.9676	0.9998
2.7713	1.32	300	1.9810	0.9998
1.3251	1.76	400	0.8096	0.6136
0.7431	2.2	500	0.6821	0.5622
0.5789	2.64	600	0.5596	0.5133
0.4866	3.08	700	0.4707	0.4381
0.3558	3.52	800	0.4653	0.4353
0.3362	3.96	900	0.4878	0.4235
0.2631	4.41	1000	0.4621	0.3907
0.2667	4.85	1100	0.4746	0.3841
0.2464	5.29	1200	0.4383	0.3780
0.205	5.73	1300	0.4207	0.3877
0.1939	6.17	1400	0.4490	0.3746
0.1644	6.61	1500	0.4325	0.3549
0.1782	7.05	1600	0.4699	0.3791

Framework versions

• Transformers 4.28.0
• Pytorch 2.0.1+cu117
• Datasets 2.14.4
• Tokenizers 0.13.3