cs2no_wav2vec2-large-xls-r-300m-czech-colab

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

• Loss: 396.8153
• Wer: 0.8457

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
3026.3663	3.51	100	472.1026	0.9873
336.2439	7.02	200	239.3806	0.9987
208.6184	10.53	300	206.7293	0.9917
182.6556	14.04	400	221.5585	0.8908
174.3151	17.54	500	262.3953	0.8921
140.57	21.05	600	225.9887	0.8330
114.5967	24.56	700	275.7823	0.8495
91.2748	28.07	800	314.0284	0.8610
80.0496	31.58	900	314.4608	0.8552
66.7338	35.09	1000	326.7965	0.8527
56.921	38.6	1100	373.0237	0.8425
50.7125	42.11	1200	374.9553	0.8527
47.4235	45.61	1300	404.8124	0.8489
45.1623	49.12	1400	396.8153	0.8457

Framework versions

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