cantillation/whisper-medium-he-teamim-silsuless-ori-TrainAndVal-Nikud

he

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

• Loss: 0.0111
• Wer: 37.4517

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: 4e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• training_steps: 1500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0155	0.02	50	0.0160	82.2919
0.0191	0.04	100	0.0271	41.2986
0.0194	0.06	150	0.0244	40.1791
0.0179	0.07	200	0.0223	34.4189
0.0157	0.09	250	0.0259	25.5445
0.016	0.11	300	0.0248	33.1773
0.0139	0.13	350	0.0214	29.3914
0.02	0.15	400	0.0223	37.3092
0.0149	0.17	450	0.0243	55.5669
0.0147	0.18	500	0.0210	70.0997
0.0134	0.2	550	0.0303	69.6519
0.0122	0.22	600	0.0182	47.2420
0.0104	0.24	650	0.0213	32.7906
0.0114	0.26	700	0.0171	25.8091
0.01	0.28	750	0.0171	40.4641
0.0071	0.3	800	0.0157	45.0641
0.0069	0.31	850	0.0172	49.5217
0.008	0.33	900	0.0169	48.7075
0.0056	0.35	950	0.0158	42.0721
0.0074	0.37	1000	0.0141	37.8587
0.0056	0.39	1050	0.0143	30.9994
0.0057	0.41	1100	0.0140	37.8995
0.0052	0.42	1150	0.0136	36.7393
0.003	0.44	1200	0.0127	34.9685
0.0034	0.46	1250	0.0119	35.5994
0.0041	0.48	1300	0.0118	37.6756
0.005	0.5	1350	0.0113	38.1641
0.0037	0.52	1400	0.0110	38.4490
0.0021	0.54	1450	0.0111	37.4517
0.0023	0.55	1500	0.0111	37.4517

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.2
• Datasets 2.16.1
• Tokenizers 0.15.0