File size: 3,101 Bytes
4ae68de |
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 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 |
---
license: apache-2.0
tags:
- automatic-speech-recognition
- librispeech_asr
- generated_from_trainer
model-index:
- name: sew-d-mid-400k-librispeech-clean-100h-ft
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. -->
# sew-d-mid-400k-librispeech-clean-100h-ft
This model is a fine-tuned version of [asapp/sew-d-mid-400k](https://huggingface.co/asapp/sew-d-mid-400k) on the LIBRISPEECH_ASR - CLEAN dataset.
It achieves the following results on the evaluation set:
- Loss: 2.3540
- Wer: 1.0536
## 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: 3e-05
- train_batch_size: 4
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 32
- total_eval_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 3.0
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 7.319 | 0.11 | 100 | 11.0572 | 1.0 |
| 3.6726 | 0.22 | 200 | 4.2003 | 1.0 |
| 2.981 | 0.34 | 300 | 3.5742 | 0.9919 |
| 2.9411 | 0.45 | 400 | 3.2599 | 1.0 |
| 2.903 | 0.56 | 500 | 2.9350 | 1.0 |
| 2.8597 | 0.67 | 600 | 2.9514 | 1.0 |
| 2.7771 | 0.78 | 700 | 2.8521 | 1.0 |
| 2.7926 | 0.9 | 800 | 2.7821 | 1.0120 |
| 2.6623 | 1.01 | 900 | 2.7027 | 0.9924 |
| 2.5893 | 1.12 | 1000 | 2.6667 | 1.0240 |
| 2.5733 | 1.23 | 1100 | 2.6341 | 1.0368 |
| 2.5455 | 1.35 | 1200 | 2.5928 | 1.0411 |
| 2.4919 | 1.46 | 1300 | 2.5695 | 1.0817 |
| 2.5182 | 1.57 | 1400 | 2.5559 | 1.1072 |
| 2.4766 | 1.68 | 1500 | 2.5229 | 1.1257 |
| 2.4267 | 1.79 | 1600 | 2.4991 | 1.1151 |
| 2.3919 | 1.91 | 1700 | 2.4768 | 1.1139 |
| 2.3883 | 2.02 | 1800 | 2.4452 | 1.0636 |
| 2.3737 | 2.13 | 1900 | 2.4304 | 1.0594 |
| 2.3569 | 2.24 | 2000 | 2.4095 | 1.0539 |
| 2.3641 | 2.35 | 2100 | 2.3997 | 1.0511 |
| 2.3281 | 2.47 | 2200 | 2.3856 | 1.0414 |
| 2.2912 | 2.58 | 2300 | 2.3750 | 1.0696 |
| 2.3028 | 2.69 | 2400 | 2.3684 | 1.0436 |
| 2.2906 | 2.8 | 2500 | 2.3613 | 1.0538 |
| 2.2822 | 2.91 | 2600 | 2.3558 | 1.0506 |
### Framework versions
- Transformers 4.12.0.dev0
- Pytorch 1.9.0+cu111
- Datasets 1.13.4.dev0
- Tokenizers 0.10.3
|