facebook/wav2vec2-xls-r-300m

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the GOOGLE/FLEURS - PS_AF dataset. It achieves the following results on the evaluation set:

  • Loss: 0.9162
  • Wer: 51.59
  • Cer: 19.72

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: 7.5e-07
  • train_batch_size: 16
  • eval_batch_size: 16
  • 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: 1000
  • training_steps: 6000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Cer Validation Loss Wer
5.0767 6.33 500 1.0 4.8783 1.0
3.1156 12.66 1000 1.0 3.0990 1.0
1.3506 18.99 1500 0.2889 1.1056 0.7031
0.9997 25.32 2000 0.2301 0.9191 0.5944
0.7838 31.65 2500 0.2152 0.8952 0.5556
0.6665 37.97 3000 0.2017 0.8908 0.5252
0.6265 44.3 3500 0.1954 0.9063 0.5133
0.5935 50.63 4000 0.1969 0.9162 0.5156
0.5174 56.96 4500 0.1972 0.9287 0.5140
0.5462 63.29 5000 0.1974 0.9370 0.5138
0.5564 69.62 5500 0.1977 0.9461 0.5148
0.5252 75.95 6000 0.9505 0.5118 0.1969

Framework versions

  • Transformers 4.26.0.dev0
  • Pytorch 1.13.1+cu117
  • Datasets 2.7.1.dev0
  • Tokenizers 0.13.2
Downloads last month
123
Safetensors
Model size
316M params
Tensor type
F32
ยท
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for ihanif/wav2vec2-xls-r-300m-pashto

Finetuned
(522)
this model

Space using ihanif/wav2vec2-xls-r-300m-pashto 1

Evaluation results