VladS159/Whisper_medium_ro_VladS_02_16_24_1500_steps_multi_gpu

Whisper Medium Ro - Sarbu Vlad

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 16.1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1719
• Wer: 12.1033

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: 16
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 3
• total_train_batch_size: 48
• total_eval_batch_size: 48
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• training_steps: 2500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1563	0.98	250	0.1542	14.6716
0.0933	1.96	500	0.1306	13.0714
0.0428	2.94	750	0.1298	11.8886
0.0243	3.92	1000	0.1353	12.0096
0.0147	4.9	1250	0.1433	12.1064
0.0083	5.88	1500	0.1572	12.2606
0.0052	6.86	1750	0.1591	12.3090
0.0037	7.84	2000	0.1665	12.0307
0.0026	8.82	2250	0.1708	12.0549
0.0021	9.8	2500	0.1719	12.1033

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.0
• Datasets 2.17.0
• Tokenizers 0.15.1