5roop/output

whisper-large-v3-mici-princ

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

• Loss: 1.4596
• Wer: 33.5008

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: 1e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• 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: 100
• training_steps: 3090
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0013	17.66	309	1.1495	37.1859
0.0009	35.31	618	1.1700	27.3032
0.0001	52.97	927	1.3428	27.7219
0.0001	70.63	1236	1.3874	27.2194
0.0001	88.29	1545	1.4141	27.3869
0.0001	105.94	1854	1.4331	33.5008
0.0001	123.6	2163	1.4445	33.3333
0.0	141.26	2472	1.4520	33.3333
0.0	158.91	2781	1.4576	33.3333
0.0	176.57	3090	1.4596	33.5008

Framework versions

• Transformers 4.38.2
• Pytorch 2.0.0+cu117
• Datasets 2.18.0
• Tokenizers 0.15.2