Makkoen/whisper-large-v3-cit-do015-wd0-lr3e-06-FULL3

./3382

This model is a fine-tuned version of openai/whisper-large-v3 on the 3382 FULL dataset. It achieves the following results on the evaluation set:

• Loss: 0.5210
• Wer Ortho: 28.1904
• Wer: 19.8548

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: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 2
• 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
0.8814	0.5249	100	0.6343	33.7614	24.7007
0.636	1.0499	200	0.5703	30.8376	22.5417
0.5203	1.5748	300	0.5425	30.0474	21.4292
0.4733	2.0997	400	0.5259	29.0695	20.3451
0.4026	2.6247	500	0.5176	28.3880	19.8548
0.3779	3.1496	600	0.5205	28.0917	19.8171
0.3365	3.6745	700	0.5146	28.2003	20.1094
0.3354	4.1995	800	0.5220	27.7558	19.6097
0.3016	4.7244	900	0.5229	28.4077	19.8171
0.2875	5.2493	1000	0.5210	28.1904	19.8548

Framework versions

• Transformers 4.44.0
• Pytorch 1.13.1+cu117
• Datasets 2.20.0
• Tokenizers 0.19.1