metadata

library_name: transformers
license: apache-2.0
base_model: openai/whisper-medium
tags:
  - generated_from_trainer
datasets:
  - bigcgen
metrics:
  - wer
model-index:
  - name: whisper-medium-bigcgen-combined-25hrs-model
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: bigcgen
          type: bigcgen
        metrics:
          - name: Wer
            type: wer
            value: 0.41335824381130315

whisper-medium-bigcgen-combined-25hrs-model

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

Loss: 0.5052
Wer: 0.4134

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: 1.75e-05
train_batch_size: 2
eval_batch_size: 2
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 8
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
training_steps: 5000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.9697	0.1219	200	0.8977	0.6518
3.2264	0.2438	400	0.7847	0.6096
2.9723	0.3657	600	0.7030	0.5395
2.9397	0.4877	800	0.6577	0.5325
2.5476	0.6096	1000	0.6216	0.4993
2.6317	0.7315	1200	0.5912	0.4900
2.4308	0.8534	1400	0.5791	0.4838
2.4843	0.9753	1600	0.5619	0.4738
1.7695	1.0969	1800	0.5714	0.4583
1.6794	1.2188	2000	0.5609	0.4493
2.0082	1.3407	2200	0.5456	0.4560
1.9617	1.4627	2400	0.5313	0.4204
1.5501	1.5846	2600	0.5262	0.4066
1.6704	1.7065	2800	0.5129	0.4163
1.5986	1.8284	3000	0.5097	0.4308
1.6337	1.9503	3200	0.5052	0.4134
0.9767	2.0719	3400	0.5273	0.4171
0.7662	2.1938	3600	0.5309	0.4162
1.0834	2.3158	3800	0.5203	0.4081

Framework versions

Transformers 4.47.1
Pytorch 2.5.1+cu124
Datasets 3.2.0
Tokenizers 0.21.0