Commit
•
0d40bab
1
Parent(s):
096481f
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- hi
|
| 4 |
+
license: apache-2.0
|
| 5 |
+
tags:
|
| 6 |
+
- automatic-speech-recognition
|
| 7 |
+
- Openslr Multilingual
|
| 8 |
+
- mozilla-foundation/common_voice_7_0
|
| 9 |
+
- generated_from_trainer
|
| 10 |
+
model-index:
|
| 11 |
+
- name: Wav2Vec2_xls_r_300m_hi_final
|
| 12 |
+
results: []
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 16 |
+
should probably proofread and complete it, then remove this comment. -->
|
| 17 |
+
|
| 18 |
+
# Wav2Vec2_xls_r_300m_hi_final
|
| 19 |
+
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the ['Openslr Multilingual and code-switching ASR challenge'](http://www.openslr.org/103/) dataset and ['mozilla-foundation/common_voice_7_0'](https://huggingface.co/datasets/mozilla-foundation/common_voice_7_0) dataset.
|
| 20 |
+
It achieves the following results on the evaluation set:
|
| 21 |
+
- Loss: 0.3035
|
| 22 |
+
- Wer: 0.3137
|
| 23 |
+
- Cer: 0.0972
|
| 24 |
+
## Model description
|
| 25 |
+
More information needed
|
| 26 |
+
## Intended uses & limitations
|
| 27 |
+
More information needed
|
| 28 |
+
## Training and evaluation data
|
| 29 |
+
More information needed
|
| 30 |
+
## Training procedure
|
| 31 |
+
### Training hyperparameters
|
| 32 |
+
The following hyperparameters were used during training:
|
| 33 |
+
- learning_rate: 0.0001
|
| 34 |
+
- train_batch_size: 16
|
| 35 |
+
- eval_batch_size: 32
|
| 36 |
+
- seed: 42
|
| 37 |
+
- gradient_accumulation_steps: 4
|
| 38 |
+
- total_train_batch_size: 64
|
| 39 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 40 |
+
- lr_scheduler_type: linear
|
| 41 |
+
- lr_scheduler_warmup_steps: 100
|
| 42 |
+
- num_epochs: 8
|
| 43 |
+
- mixed_precision_training: Native AMP
|
| 44 |
+
### Training results
|
| 45 |
+
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|
| 46 |
+
|:-------------:|:-----:|:----:|:---------------:|:------:|:------:|
|
| 47 |
+
| 0.9821 | 0.64 | 400 | 0.5059 | 0.4783 | 0.1573 |
|
| 48 |
+
| 0.6861 | 1.28 | 800 | 0.4201 | 0.4247 | 0.1356 |
|
| 49 |
+
| 0.585 | 1.92 | 1200 | 0.3797 | 0.3811 | 0.1210 |
|
| 50 |
+
| 0.5193 | 2.56 | 1600 | 0.3577 | 0.3652 | 0.1152 |
|
| 51 |
+
| 0.4583 | 3.21 | 2000 | 0.3422 | 0.3519 | 0.1111 |
|
| 52 |
+
| 0.4282 | 3.85 | 2400 | 0.3261 | 0.3450 | 0.1071 |
|
| 53 |
+
| 0.3951 | 4.49 | 2800 | 0.3201 | 0.3325 | 0.1048 |
|
| 54 |
+
| 0.3619 | 5.13 | 3200 | 0.3167 | 0.3296 | 0.1030 |
|
| 55 |
+
| 0.345 | 5.77 | 3600 | 0.3157 | 0.3210 | 0.1013 |
|
| 56 |
+
| 0.338 | 6.41 | 4000 | 0.3051 | 0.3143 | 0.0982 |
|
| 57 |
+
| 0.3155 | 7.05 | 4400 | 0.3059 | 0.3154 | 0.0986 |
|
| 58 |
+
| 0.3057 | 7.69 | 4800 | 0.3035 | 0.3137 | 0.0972 |
|
| 59 |
+
### Framework versions
|
| 60 |
+
- Transformers 4.17.0.dev0
|
| 61 |
+
- Pytorch 1.10.2+cu102
|
| 62 |
+
- Datasets 1.18.3
|
| 63 |
+
- Tokenizers 0.11.0
|