VladS159/Whisper_medium_ro_VladS_2000_steps_multi_gpu_23_02_2024

Whisper Medium Ro - Sarbu Vlad - multi gpu

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

• Loss: 0.1620
• Wer: 12.1820

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: 200
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.148	0.98	250	0.1494	14.1574
0.0875	1.96	500	0.1295	12.9080
0.0404	2.94	750	0.1285	11.8734
0.0227	3.92	1000	0.1353	12.1094
0.0139	4.9	1250	0.1409	11.9702
0.0076	5.88	1500	0.1539	12.0459
0.005	6.86	1750	0.1599	12.1880
0.0039	7.84	2000	0.1620	12.1820

Framework versions

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