Makkoen/whisper-large-v3-cit-do01-wd0-lr3e-06-FULL4c

./4585

This model is a fine-tuned version of openai/whisper-large-v3 on the 4585 FULL-2024-09-26 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4883
• Wer Ortho: 27.5525
• Wer: 19.6598

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: 1200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.7769	0.7752	200	0.5669	31.2018	22.9917
0.5359	1.5504	400	0.5151	29.0481	20.9467
0.4524	2.3256	600	0.4949	28.1973	20.0166
0.3889	3.1008	800	0.4895	27.6788	19.6471
0.3431	3.8760	1000	0.4841	27.4063	19.4368
0.3196	4.6512	1200	0.4883	27.5525	19.6598

Framework versions

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