Whisper largeV2 Italian MLS
This model is a fine-tuned version of openai/whisper-large-v2 on the facebook/multilingual_librispeech italian dataset. It achieves the following results on the evaluation set:
- Loss: 0.2051
- Wer: 8.3353
Model description
The model is fine-tuned for 4000 updates/steps on multilingual librispeech Italian train data.
- Zero-shot - 13.8 (MLS Italian test)
- Fine-tune MLS Italian train - 8.33 (MLS Italian test) (-40%)
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: 32
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 4000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.1115 | 1.02 | 1000 | 0.2116 | 9.4217 |
| 0.0867 | 2.03 | 2000 | 0.1964 | 9.7823 |
| 0.0447 | 3.05 | 3000 | 0.2001 | 9.6409 |
| 0.0426 | 4.07 | 4000 | 0.2051 | 8.3353 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.0+cu117
- Datasets 2.7.1.dev0
- Tokenizers 0.13.2
- Downloads last month
- 11
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Dataset used to train sgangireddy/whisper-largev2-mls-it
Evaluation results
- Wer on facebook/multilingual_librispeech italiantest set self-reported8.335