Makkoen/whisper-large-v3-cit-do005-wd0-lr5e-06-steps2000-FULL6b

FULL6

This model is a fine-tuned version of openai/whisper-large-v3 on the 9651 FULL-2024-12-01 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3740
• Wer Ortho: 20.5948
• Wer: 14.5676

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: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.646	0.3683	200	0.4534	25.1768	18.6441
0.5098	0.7366	400	0.4155	24.7355	18.4413
0.4491	1.1050	600	0.3936	21.8793	15.6599
0.3674	1.4733	800	0.3875	21.7736	15.4860
0.365	1.8416	1000	0.3765	21.2869	15.1644
0.3171	2.2099	1200	0.3779	21.1358	15.0978
0.2886	2.5783	1400	0.3755	20.7610	14.6661
0.2871	2.9466	1600	0.3689	20.6009	14.6023
0.2395	3.3149	1800	0.3752	20.6825	14.8110
0.2444	3.6832	2000	0.3740	20.5948	14.5676

Framework versions

• Transformers 4.44.0
• Pytorch 1.13.1+cu117
• Datasets 2.21.0
• Tokenizers 0.19.1