wav2vec2-large-xls-r-300m-slowenian-with-lm

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3935
• Wer: 0.3480

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: 0.0001
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.9937	2.5	100	3.1565	1.0
3.0466	5.0	200	3.0009	0.9992
2.9708	7.5	300	2.9494	0.9992
2.0519	10.0	400	0.8874	0.7290
0.5773	12.5	500	0.5258	0.5037
0.3427	15.0	600	0.4767	0.4649
0.2612	17.5	700	0.4549	0.4209
0.212	20.0	800	0.4294	0.3860
0.1748	22.5	900	0.4085	0.3769
0.1587	25.0	1000	0.4017	0.3673
0.1435	27.5	1100	0.3927	0.3538
0.1314	30.0	1200	0.3935	0.3480

Framework versions

• Transformers 4.17.0
• Pytorch 1.9.0+cu111
• Datasets 1.18.4
• Tokenizers 0.11.6