Makkoen/whisper-large-v3-cit-do01-wd0-lr3e-06-steps1400-FULL6

./9712

This model is a fine-tuned version of openai/whisper-large-v3 on the 9712 FULL-2024-10-24 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3774
• Wer Ortho: 21.1633
• Wer: 15.3466

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: 3e-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 300
• training_steps: 1400
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.6939	0.3661	200	0.4724	26.4867	19.6596
0.5324	0.7323	400	0.4202	23.7327	17.6207
0.4623	1.0984	600	0.3970	21.8897	16.1424
0.4049	1.4645	800	0.3879	22.1227	16.3051
0.3962	1.8307	1000	0.3811	21.0907	15.3844
0.377	2.1968	1200	0.3787	21.3722	15.6109
0.3422	2.5629	1400	0.3774	21.1633	15.3466

Framework versions

• Transformers 4.45.1
• Pytorch 1.13.1+cu117
• Datasets 3.0.1
• Tokenizers 0.20.0