Kushtrim/whisper-large-v3-turbo-sq-test2

Whisper Large V3 Turbo SQ

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the Common Voice 19.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3040
• Wer: 22.4519

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: 500
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5578	0.3163	250	0.6037	43.1130
0.4707	0.6325	500	0.5295	40.3256
0.4221	0.9488	750	0.4559	35.2985
0.2853	1.2650	1000	0.4205	33.3103
0.2685	1.5813	1250	0.3798	30.7844
0.256	1.8975	1500	0.3552	28.5890
0.18	2.2138	1750	0.3480	27.5728
0.2158	2.5300	2000	0.3349	27.2521
0.1396	2.8463	2250	0.3182	24.2526
0.1123	3.1626	2500	0.3175	23.5520
0.124	3.4788	2750	0.3100	23.4090
0.0908	3.7951	3000	0.3040	22.4519

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1