OverloadedOperator/tokomni-whisper-i3

TK_Whisper_ASR

This model is a fine-tuned version of openai/whisper-tiny on the KAN Hebrew Speech dataset. It achieves the following results on the evaluation set:

• Loss: 5.4256

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
No log	1.0	1	6.4006
No log	2.0	2	6.4006
No log	3.0	3	6.4006
No log	4.0	4	6.4006
No log	5.0	5	6.4006
No log	6.0	6	6.3966
No log	7.0	7	6.3910
No log	8.0	8	6.3844
No log	9.0	9	6.3857
No log	10.0	10	6.3694
No log	11.0	11	6.3381
No log	12.0	12	6.3115
No log	13.0	13	6.2788
No log	14.0	14	6.2411
No log	15.0	15	6.1932
No log	16.0	16	6.1552
No log	17.0	17	6.1146
No log	18.0	18	6.0711
No log	19.0	19	6.0122
No log	20.0	20	5.9643
No log	21.0	21	5.9193
No log	22.0	22	5.8783
No log	23.0	23	5.8258
No log	24.0	24	5.7742
6.7306	25.0	25	5.7174
6.7306	26.0	26	5.6629
6.7306	27.0	27	5.6041
6.7306	28.0	28	5.5453
6.7306	29.0	29	5.4896
6.7306	30.0	30	5.4256

Framework versions

• PEFT 0.9.0
• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.0