Makkoen/whisper-large-v3-cit-do015-wd0-lr1e-06-FULL4

./4607

This model is a fine-tuned version of openai/whisper-large-v3 on the 4607 FULL-2024-09-26 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5059
• Wer Ortho: 28.4797
• Wer: 21.0751

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: 1e-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
0.9206	0.7715	200	0.6309	33.9104	25.9900
0.6533	1.5429	400	0.5581	30.2736	22.5910
0.5875	2.3144	600	0.5322	29.5128	22.8648
0.5351	3.0858	800	0.5176	29.3103	21.8431
0.5126	3.8573	1000	0.5112	28.7100	21.3222
0.4956	4.6287	1200	0.5063	28.6053	21.0751
0.4785	5.4002	1400	0.5059	28.4797	21.0751

Framework versions

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