Makkoen/whisper-large-v3-cit-do015-wd0-lr3e-06-FULL3b

./3479

This model is a fine-tuned version of openai/whisper-large-v3 on the 3479 clips dataset. It achieves the following results on the evaluation set:

• Loss: 0.5117
• Wer Ortho: 27.4535
• Wer: 19.3463

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: 200
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.8906	0.5109	100	0.6318	33.6218	25.1010
0.6428	1.0217	200	0.5620	30.8415	22.5971
0.5279	1.5326	300	0.5435	32.0107	23.8886
0.4958	2.0434	400	0.5244	30.0037	21.7800
0.4238	2.5543	500	0.5171	28.4662	20.2337
0.4016	3.0651	600	0.5132	28.0980	19.8647
0.3562	3.5760	700	0.5132	27.6100	19.7505
0.3467	4.0868	800	0.5103	27.1037	19.0828
0.308	4.5977	900	0.5117	27.3246	19.1618
0.3174	5.1086	1000	0.5117	27.4535	19.3463

Framework versions

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