|
--- |
|
license: apache-2.0 |
|
tags: |
|
- automatic-speech-recognition |
|
- librispeech_asr |
|
- generated_from_trainer |
|
model-index: |
|
- name: sew-mid-100k-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-mid-100k-librispeech-clean-100h-ft |
|
|
|
This model is a fine-tuned version of [asapp/sew-mid-100k](https://huggingface.co/asapp/sew-mid-100k) on the LIBRISPEECH_ASR - CLEAN dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.1976 |
|
- Wer: 0.1665 |
|
|
|
## 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 | |
|
|:-------------:|:-----:|:----:|:---------------:|:------:| |
|
| 3.4274 | 0.11 | 100 | 4.1419 | 1.0 | |
|
| 2.9657 | 0.22 | 200 | 3.1203 | 1.0 | |
|
| 2.9069 | 0.34 | 300 | 3.0107 | 1.0 | |
|
| 2.8666 | 0.45 | 400 | 2.8960 | 1.0 | |
|
| 1.4535 | 0.56 | 500 | 1.4062 | 0.8664 | |
|
| 0.6821 | 0.67 | 600 | 0.5530 | 0.4930 | |
|
| 0.4827 | 0.78 | 700 | 0.4122 | 0.3630 | |
|
| 0.4485 | 0.9 | 800 | 0.3597 | 0.3243 | |
|
| 0.2666 | 1.01 | 900 | 0.3104 | 0.2790 | |
|
| 0.2378 | 1.12 | 1000 | 0.2913 | 0.2613 | |
|
| 0.2516 | 1.23 | 1100 | 0.2702 | 0.2452 | |
|
| 0.2456 | 1.35 | 1200 | 0.2619 | 0.2338 | |
|
| 0.2392 | 1.46 | 1300 | 0.2466 | 0.2195 | |
|
| 0.2117 | 1.57 | 1400 | 0.2379 | 0.2092 | |
|
| 0.1837 | 1.68 | 1500 | 0.2295 | 0.2029 | |
|
| 0.1757 | 1.79 | 1600 | 0.2240 | 0.1949 | |
|
| 0.1626 | 1.91 | 1700 | 0.2195 | 0.1927 | |
|
| 0.168 | 2.02 | 1800 | 0.2137 | 0.1853 | |
|
| 0.168 | 2.13 | 1900 | 0.2123 | 0.1839 | |
|
| 0.1576 | 2.24 | 2000 | 0.2095 | 0.1803 | |
|
| 0.1756 | 2.35 | 2100 | 0.2075 | 0.1776 | |
|
| 0.1467 | 2.47 | 2200 | 0.2049 | 0.1754 | |
|
| 0.1702 | 2.58 | 2300 | 0.2013 | 0.1722 | |
|
| 0.177 | 2.69 | 2400 | 0.1993 | 0.1701 | |
|
| 0.1417 | 2.8 | 2500 | 0.1983 | 0.1688 | |
|
| 0.1302 | 2.91 | 2600 | 0.1977 | 0.1678 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.12.0.dev0 |
|
- Pytorch 1.9.0+cu111 |
|
- Datasets 1.13.4.dev0 |
|
- Tokenizers 0.10.3 |
|
|