whisper-large-v3-turbo-ivrit-ai-coursera-fine-tuned

This model is a fine-tuned version of ivrit-ai/whisper-large-v3-turbo on the dataset imvladikon/hebrew_speech_coursera. It achieves the following results on the evaluation set:

• Loss: 0.2829

Model description

This model created for my work for the Open University Of Israel. Here you can see the notebook that used to create this model, and here you can find me displaying the notebook. I think that this model is useless becaus it has lower performance from its base model.

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: 5
• eval_batch_size: 5
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_ratio: 0.1
• lr_scheduler_warmup_steps: 500
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss
0.1907	0.1641	500	0.2266
0.2283	0.3283	1000	0.2217
0.2253	0.4924	1500	0.2154
0.2257	0.6566	2000	0.2080
0.2138	0.8207	2500	0.2102
0.2153	0.9849	3000	0.2056
0.1615	1.1490	3500	0.2128
0.1588	1.3132	4000	0.1677
0.1628	1.4773	4500	0.1656
0.168	1.6415	5000	0.1798
0.167	1.8056	5500	0.1710
0.1663	1.9698	6000	0.1828
0.1297	2.1339	6500	0.1722
0.1196	2.2981	7000	0.1762
0.1336	2.4622	7500	0.1779
0.1258	2.6264	8000	0.1821
0.1275	2.7905	8500	0.1796
0.1331	2.9547	9000	0.1786
0.0988	3.1188	9500	0.1982
0.0933	3.2830	10000	0.1888
0.0963	3.4471	10500	0.1927
0.0946	3.6113	11000	0.1979
0.1018	3.7754	11500	0.2031
0.1027	3.9396	12000	0.1971
0.0795	4.1037	12500	0.2016
0.0698	4.2679	13000	0.2017
0.0736	4.4320	13500	0.2058
0.0747	4.5962	14000	0.2033
0.0768	4.7603	14500	0.2057
0.0801	4.9245	15000	0.2076
0.067	5.0886	15500	0.2196
0.0539	5.2528	16000	0.2185
0.0563	5.4169	16500	0.2220
0.0594	5.5811	17000	0.2265
0.0651	5.7452	17500	0.2176
0.0655	5.9094	18000	0.2227
0.0533	6.0735	18500	0.2387
0.0441	6.2377	19000	0.2334
0.0474	6.4018	19500	0.2343
0.0506	6.5660	20000	0.2387
0.0504	6.7301	20500	0.2373
0.0502	6.8943	21000	0.2318
0.0441	7.0584	21500	0.2524
0.0375	7.2226	22000	0.2533
0.0379	7.3867	22500	0.2491
0.0382	7.5509	23000	0.2635
0.0427	7.7150	23500	0.2506
0.0439	7.8792	24000	0.2430
0.043	8.0433	24500	0.2575
0.0296	8.2075	25000	0.2617
0.0309	8.3716	25500	0.2797
0.0366	8.5358	26000	0.2689
0.0351	8.6999	26500	0.2687
0.0384	8.8641	27000	0.2643
0.0365	9.0282	27500	0.2688
0.0265	9.1924	28000	0.2903
0.0299	9.3565	28500	0.2742
0.0347	9.5207	29000	0.2754
0.0311	9.6848	29500	0.2744
0.0345	9.8490	30000	0.2829

Framework versions

• Transformers 4.48.1
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1