metadata

base_model: facebook/wav2vec2-large-xlsr-53
datasets:
  - common_voice_17_0
license: apache-2.0
metrics:
  - wer
tags:
  - generated_from_trainer
model-index:
  - name: wav2vec2-large-xlsr-Mongolian-cv17-base
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: common_voice_17_0
          type: common_voice_17_0
          config: mn
          split: validation
          args: mn
        metrics:
          - type: wer
            value: 0.7902951968892054
            name: Wer

wav2vec2-large-xlsr-Mongolian-cv17-base

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

Loss: 0.9632
Wer: 0.7903

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: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 20
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.1940	40	13.1486	1.0009
No log	2.3881	80	7.6639	1.0
No log	3.5821	120	3.4345	1.0
No log	4.7761	160	3.1527	1.0
No log	5.9701	200	3.1223	1.0
No log	7.1642	240	3.1137	1.0
No log	8.3582	280	3.1017	1.0
No log	9.5522	320	3.0909	1.0
No log	10.7463	360	3.0363	1.0
5.1112	11.9403	400	2.8364	1.0
5.1112	13.1343	440	2.0134	1.0078
5.1112	14.3284	480	1.3866	1.0511
5.1112	15.5224	520	1.1292	0.9320
5.1112	16.7164	560	1.0117	0.9017
5.1112	17.9104	600	0.9756	0.8339
5.1112	19.1045	640	0.9632	0.7903

Framework versions

Transformers 4.42.4
Pytorch 2.3.1+cu121
Datasets 2.20.0
Tokenizers 0.19.1