|
--- |
|
language: |
|
- hi |
|
license: apache-2.0 |
|
tags: |
|
- hf-asr-leaderboard |
|
- generated_from_trainer |
|
datasets: |
|
- mozilla-foundation/common_voice_11_0 |
|
model-index: |
|
- name: Whisper Small Hi - Sanchit Gandhi |
|
results: [] |
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# Whisper Small Hi - Sanchit Gandhi |
|
|
|
This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the Common Voice 11.0 dataset. |
|
It achieves the following results on the evaluation set: |
|
- eval_loss: 0.2226 |
|
- eval_wer: 10.0023 |
|
- eval_runtime: 65.2041 |
|
- eval_samples_per_second: 1.748 |
|
- eval_steps_per_second: 0.23 |
|
- epoch: 19.51 |
|
- step: 800 |
|
|
|
## 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: 200 |
|
- training_steps: 1000 |
|
- mixed_precision_training: Native AMP |
|
|
|
### Framework versions |
|
|
|
- Transformers 4.25.0.dev0 |
|
- Pytorch 1.12.1+cu113 |
|
- Datasets 2.7.1 |
|
- Tokenizers 0.13.2 |
|
|
