fb-youtube-vi-large

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the PHONGDTD/YOUTUBE_CASUAL_AUDIO - NA dataset.

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: 2e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 8
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• num_epochs: 25.0
• mixed_precision_training: Native AMP

Training results

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.0+cu111
• Datasets 1.18.3
• Tokenizers 0.10.3