VladS159/Whisper_medium_ro_VladS_10000_steps_3_gpus_small_lr

Whisper Medium Ro - Sarbu Vlad - 3 gpus

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 17.0 + Romanian speech synthesis dataset. It achieves the following results on the evaluation set:

• Loss: 0.0878
• Wer: 6.3598

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: 5e-06
• train_batch_size: 10
• eval_batch_size: 10
• seed: 42
• distributed_type: multi-GPU
• num_devices: 3
• total_train_batch_size: 30
• total_eval_batch_size: 30
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2346	0.42	500	0.1671	15.7382
0.1441	0.85	1000	0.1183	12.1318
0.0887	1.27	1500	0.1026	10.3012
0.0821	1.7	2000	0.0926	9.8230
0.0403	2.12	2500	0.0871	9.1590
0.0394	2.55	3000	0.0914	8.8788
0.0381	2.97	3500	0.0790	8.2909
0.0178	3.4	4000	0.0825	7.7579
0.0185	3.82	4500	0.0776	7.4929
0.0102	4.25	5000	0.0818	7.4350
0.0105	4.67	5500	0.0803	6.9599
0.0054	5.1	6000	0.0830	6.9264
0.0046	5.52	6500	0.0827	6.6949
0.005	5.95	7000	0.0831	6.7101
0.0043	6.37	7500	0.0840	6.6462
0.003	6.8	8000	0.0845	6.5274
0.0017	7.22	8500	0.0875	6.5121
0.0015	7.65	9000	0.0867	6.4025
0.0012	8.07	9500	0.0875	6.3568
0.001	8.5	10000	0.0878	6.3598

Framework versions

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