Makkoen/whisper-large-v3-cit-do00-wd0-lr5e-06-steps1600-FULL6

FULL6

This model is a fine-tuned version of openai/whisper-large-v3 on the 9712 FULL-2024-11-28 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3640
• Wer Ortho: 20.3069
• Wer: 14.6031

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-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: 1600
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.6552	0.3661	200	0.4502	25.3125	18.9306
0.5133	0.7323	400	0.4050	22.6977	16.4939
0.4417	1.0984	600	0.3848	21.6687	15.9014
0.3677	1.4645	800	0.3760	22.0198	16.0175
0.3611	1.8307	1000	0.3687	21.1089	15.1927
0.329	2.1968	1200	0.3691	20.9636	15.1462
0.2864	2.5629	1400	0.3657	20.4915	14.7512
0.2842	2.9291	1600	0.3640	20.3069	14.6031

Framework versions

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