File size: 1,738 Bytes
6eb26f4 aa60d09 6eb26f4 aa60d09 6eb26f4 aa60d09 6eb26f4 aa60d09 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 |
---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- common_voice
model-index:
- name: ArabicSpeechToText
results: []
language:
- ar
---
<!-- 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. -->
# ArabicSpeechToText
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the common_voice dataset.
It achieves the following results on the evaluation set:
- Loss: 0.3207
- Wer: 0.2901
## 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: 0.0003
- train_batch_size: 2
- eval_batch_size: 8
- seed: 8
- gradient_accumulation_steps: 8
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 3
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 5.6589 | 0.52 | 700 | 0.8361 | 0.7598 |
| 0.7367 | 1.03 | 1400 | 0.5243 | 0.4771 |
| 0.479 | 1.55 | 2100 | 0.4025 | 0.3756 |
| 0.3742 | 2.06 | 2800 | 0.3651 | 0.3272 |
| 0.2846 | 2.58 | 3500 | 0.3207 | 0.2901 |
### Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu113
- Datasets 2.9.0
- Tokenizers 0.10.3 |