wav2vec2-large-xls-r-300m-saha-yakut

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6404
• Wer: 0.5217

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: 0.0002
• train_batch_size: 12
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 24
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	3.4483	150	3.1675	1.0
No log	6.8966	300	1.1392	0.8509
3.5746	10.3448	450	0.6103	0.5812
3.5746	13.7931	600	0.6152	0.5565
3.5746	17.2414	750	0.6420	0.5382
0.2138	20.6897	900	0.6344	0.5245
0.2138	24.1379	1050	0.6543	0.5303
0.1148	27.5862	1200	0.6404	0.5217

Framework versions

• Transformers 4.41.0.dev0
• Pytorch 2.2.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1