metadata

language:
  - hi
license: apache-2.0
base_model: openai/whisper-medium
tags:
  - generated_from_trainer
datasets:
  - velocity-whisper-tiny
metrics:
  - wer
model-index:
  - name: whisper-tiny-finetuned-hinglish
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: whisper-training
          type: velocity-whisper-tiny
          args: 'config: hi, split: test'
        metrics:
          - name: Wer
            type: wer
            value: 42.262816735415434

whisper-tiny-finetuned-hinglish

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

Loss: 0.7758
Wer: 42.2628

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
num_epochs: 40
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3632	1.7825	1000	0.3962	51.0784
0.2411	3.5651	2000	0.3428	45.1149
0.1242	5.3476	3000	0.3459	42.1685
0.0813	7.1301	4000	0.3610	42.1685
0.0654	8.9127	5000	0.3949	41.9210
0.0309	10.6952	6000	0.4422	42.7814
0.0161	12.4777	7000	0.4836	42.3925
0.0067	14.2602	8000	0.5291	42.9346
0.0032	16.0428	9000	0.5645	42.4514
0.0031	17.8253	10000	0.5951	42.7814
0.002	19.6078	11000	0.6248	42.5103
0.0007	21.3904	12000	0.6486	42.8167
0.0004	23.1729	13000	0.6760	42.0625
0.0008	24.9554	14000	0.6982	42.4396
0.0018	26.7380	15000	0.7149	42.4985
0.0002	28.5205	16000	0.7172	41.8739
0.0001	30.3030	17000	0.7307	42.4042
0.0001	32.0856	18000	0.7399	42.0742
0.0001	33.8681	19000	0.7497	42.1332
0.0001	35.6506	20000	0.7608	42.0860
0.0	37.4332	21000	0.7695	41.9682
0.0	39.2157	22000	0.7758	42.2628

Framework versions

Transformers 4.41.2
Pytorch 2.3.1+cu121
Datasets 2.19.2
Tokenizers 0.19.1