wav2vec2-large-xls-r-300m-bulgarian

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - BG dataset. It achieves the following results on the evaluation set:

• Loss: 0.4487
• Wer: 0.4674

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: 7e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.9774	6.33	500	2.9769	1.0
1.3453	12.66	1000	0.6523	0.6980
1.1658	18.99	1500	0.5636	0.6359
1.0797	25.32	2000	0.5004	0.5759
1.044	31.65	2500	0.4958	0.5569
0.9915	37.97	3000	0.4971	0.5350
0.9429	44.3	3500	0.4829	0.5229
0.9266	50.63	4000	0.4515	0.5074
0.8965	56.96	4500	0.4599	0.5039
0.878	63.29	5000	0.4735	0.4954
0.8494	69.62	5500	0.4460	0.4878
0.8343	75.95	6000	0.4510	0.4795
0.8236	82.28	6500	0.4538	0.4789
0.8069	88.61	7000	0.4526	0.4748
0.7958	94.94	7500	0.4496	0.4700

Framework versions

• Transformers 4.16.0.dev0
• Pytorch 1.10.1+cu102
• Datasets 1.17.1.dev0
• Tokenizers 0.11.0