README.md

metadata

language:
  - he
license: apache-2.0
base_model: openai/whisper-small
tags:
  - hf-asr-leaderboard
  - generated_from_trainer
datasets:
  - ivrit-ai/whisper-training
metrics:
  - wer
model-index:
  - name: Whisper Small Hebrew
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: ivrit-ai/whisper-training
          type: ivrit-ai/whisper-training
          args: 'config: he, split: train'
        metrics:
          - name: Wer
            type: wer
            value: 37.8652

Whisper Small Hebrew

This model is a fine-tuned version of openai/whisper-small on the ivrit-ai/whisper-training dataset. It achieves the following results on the evaluation set:

• Loss: 0.4241
• Wer: 37.8652

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4946	0.13	500	0.4572	46.4463
0.4629	0.25	1000	0.4492	43.7663
0.4067	0.38	1500	0.4337	42.6317
0.3663	0.5	2000	0.3892	41.8427
0.3857	0.63	2500	0.4017	40.7473
0.3795	0.75	3000	0.4011	39.4823
0.368	0.88	3500	0.3967	39.9778
0.2353	1.01	4000	0.3801	38.3281
0.2405	1.13	4500	0.4062	41.5428
0.2512	1.26	5000	0.3975	38.6215
0.2433	1.38	5500	0.4035	38.5824
0.2368	1.51	6000	0.3983	37.8652
0.2592	1.63	6500	0.4184	39.1041
0.2629	1.76	7000	0.4000	39.5475
0.2318	1.88	7500	0.4012	39.1954
0.1658	2.01	8000	0.3941	39.1367
0.1546	2.14	8500	0.4226	39.8865
0.1665	2.26	9000	0.4295	40.9755
0.1642	2.39	9500	0.4314	41.1255
0.1694	2.51	10000	0.4241	40.6755

Framework versions

• Transformers 4.40.0.dev0
• Pytorch 2.2.1
• Datasets 2.18.0
• Tokenizers 0.15.2