anton-l's picture
anton-l HF staff
Upload README.md
00fd196
---
language:
- lt
license: apache-2.0
tags:
- automatic-speech-recognition
- mozilla-foundation/common_voice_7_0
- generated_from_trainer
- lt
- robust-speech-event
- model_for_talk
- hf-asr-leaderboard
datasets:
- mozilla-foundation/common_voice_7_0
model-index:
- name: XLS-R-300M - Lithuanian
results:
- task:
name: Automatic Speech Recognition
type: automatic-speech-recognition
dataset:
name: Common Voice 7
type: mozilla-foundation/common_voice_7_0
args: lt
metrics:
- name: Test WER
type: wer
value: 24.859
- name: Test CER
type: cer
value: 4.764
---
<!-- 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. -->
# wav2vec2-large-xls-r-300m-lithuanian
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - LT dataset.
It achieves the following results on the evaluation set:
- Loss: 0.1722
- Wer: 0.2486
## 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: 7e-05
- train_batch_size: 32
- eval_batch_size: 1
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2000
- num_epochs: 50.0
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:-----:|:---------------:|:------:|
| 1.6837 | 8.0 | 2000 | 0.6649 | 0.7515 |
| 1.1105 | 16.0 | 4000 | 0.2386 | 0.3436 |
| 1.0069 | 24.0 | 6000 | 0.2008 | 0.2968 |
| 0.9417 | 32.0 | 8000 | 0.1915 | 0.2774 |
| 0.887 | 40.0 | 10000 | 0.1819 | 0.2616 |
| 0.8563 | 48.0 | 12000 | 0.1729 | 0.2475 |
### Framework versions
- Transformers 4.16.0.dev0
- Pytorch 1.10.1+cu102
- Datasets 1.17.1.dev0
- Tokenizers 0.11.0