Makkoen/whisper-large-v3-cit-do01-wd0-lr5e-06-steps1600-FULL6

FULL6

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: 0.3645
• Wer Ortho: 20.3644
• Wer: 14.6990

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
0.6529	0.3661	200	0.4500	25.1823	18.8290
0.5129	0.7323	400	0.4050	22.7643	16.5316
0.4413	1.0984	600	0.3850	22.1197	16.4532
0.3677	1.4645	800	0.3770	21.8382	15.6719
0.3614	1.8307	1000	0.3691	21.0786	15.2740
0.3297	2.1968	1200	0.3696	20.9031	15.1201
0.2872	2.5629	1400	0.3660	20.5248	14.8442
0.2849	2.9291	1600	0.3645	20.3644	14.6990

Framework versions

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