dgx2_whisper_small_mozilla_noisy_distil_epochs_50_batch_8

This model is a fine-tuned version of rohitp1/kkkh_whisper_small_distillation_att_loss_mozilla_epochs_100_batch_4_concat_dataset on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.1254
• Wer: 20.5209

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 256
• total_train_batch_size: 2048
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine_with_restarts
• lr_scheduler_warmup_ratio: 0.2
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0006	4.39	150	0.5538	19.3893
0.1453	8.78	300	0.7751	20.7263
0.3233	13.17	450	0.8857	20.7994
0.486	17.55	600	1.0980	20.6462
0.6433	21.94	750	1.1264	20.5835
0.6452	26.33	900	1.1254	20.5209

Framework versions

• Transformers 4.28.1
• Pytorch 1.12.1
• Datasets 2.8.0
• Tokenizers 0.13.2