wav2vec2-large-xls-r-300m-czech-colab-finetuned

This model is a fine-tuned version of facebook/wav2vec2-lv-60-espeak-cv-ft on the voxpopuli dataset. It achieves the following results on the evaluation set:

• Loss: 624.5939
• Wer: 0.6178

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3007.3212	3.51	100	1006.7374	0.9865
354.3011	7.02	200	563.6080	0.9980
211.5289	10.53	300	599.5796	0.9165
187.8653	14.04	400	447.1478	0.8099
163.1056	17.54	500	430.5204	0.6875
143.0342	21.05	600	413.8947	0.6850
116.0388	24.56	700	435.5743	0.6737
95.5554	28.07	800	490.6329	0.6339
80.6966	31.58	900	493.9658	0.6344
68.7335	35.09	1000	525.7507	0.6263
58.3269	38.6	1100	582.5747	0.6128
54.3181	42.11	1200	600.8087	0.6308
48.5287	45.61	1300	594.6959	0.6112
43.041	49.12	1400	624.5939	0.6178

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0