Beta
Explore
Marketplace
Neural Labs
Chat
Wallet
Docs
whisper_3 – AI Model by Zipei-KTH | AlphaNeural AI
You can deploy this model and start earning money today!
Zipei-KTH
/
whisper_3
like
0
transformers
tensorboard
safetensors
whisper
automatic-speech-recognition
generated_from_trainer
openai/whisper-small
finetune
apache-2.0
endpoints_compatible
us
Views
No views yet
Model card
Files and Versions
Community
API
Deploy
whisper_3
This model is a fine-tuned version of
openai/whisper-small
on the None dataset. It achieves the following results on the evaluation set:
Loss: 0.2148
Wer: 118.3054
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: 8
eval_batch_size: 2
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
training_steps: 3100
mixed_precision_training: Native AMP
Training results
Training Loss
Epoch
Step
Validation Loss
Wer
0.0353
2.84
1000
0.1834
58.2714
0.0024
5.67
2000
0.2006
101.7201
0.0008
8.51
3000
0.2148
118.3054
Framework versions
Transformers 4.35.2
Pytorch 2.1.1+cu121
Datasets 2.15.0
Tokenizers 0.15.0
