Makkoen/whisper-large-v3-cit-do01-wd0-lr3e-06-steps1200-FULL4test

./4528

This model is a fine-tuned version of openai/whisper-large-v3 on the 4528 FULL-2024-10-24 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5039
• Wer Ortho: 28.2453
• Wer: 20.5359

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
0.7791	0.7851	200	0.5708	32.0818	24.2128
0.5319	1.5702	400	0.5254	30.1601	22.5438
0.4597	2.3553	600	0.5083	28.4089	21.1357
0.3953	3.1403	800	0.5049	28.2658	20.6467
0.3522	3.9254	1000	0.4995	28.1840	20.4577
0.3191	4.7105	1200	0.5039	28.2453	20.5359

Framework versions

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