imvladikon's picture
Librarian Bot: Add base_model information to model (#3)
d67afbc
metadata
language:
  - he
license: apache-2.0
tags:
  - automatic-speech-recognition
  - robust-speech-event
  - he
  - generated_from_trainer
  - hf-asr-leaderboard
base_model: facebook/wav2vec2-xls-r-1b
model-index:
  - name: wav2vec2-xls-r-1b-hebrew
    results: []

wav2vec2-xls-r-1b-hebrew

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3533
  • Wer: 0.2251

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: 6
  • eval_batch_size: 6
  • seed: 42
  • distributed_type: multi-GPU
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 24
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 400
  • num_epochs: 20.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
3.3587 0.47 400 1.1883 0.8392
1.8377 0.95 800 0.8831 0.6852
1.7118 1.42 1200 0.8031 0.6566
1.6741 1.89 1600 0.7518 0.6104
1.6163 2.36 2000 0.6888 0.5591
1.5782 2.84 2400 0.6580 0.5165
1.5548 3.31 2800 0.6506 0.5184
1.5249 3.78 3200 0.6198 0.5028
1.5078 4.26 3600 0.5992 0.4932
1.4836 4.73 4000 0.5705 0.4651
1.4505 5.2 4400 0.5489 0.4508
1.4481 5.67 4800 0.5577 0.4562
1.4136 6.15 5200 0.5452 0.4371
1.3861 6.62 5600 0.5101 0.4087
1.3772 7.09 6000 0.4933 0.3951
1.3478 7.56 6400 0.4849 0.3922
1.3394 8.04 6800 0.4805 0.3892
1.3095 8.51 7200 0.4839 0.3834
1.306 8.98 7600 0.4611 0.3587
1.2707 9.46 8000 0.4545 0.3730
1.2626 9.93 8400 0.4516 0.3524
1.2412 10.4 8800 0.4314 0.3310
1.2456 10.87 9200 0.4401 0.3459
1.2081 11.35 9600 0.4399 0.3356
1.1998 11.82 10000 0.4195 0.3215
1.1826 12.29 10400 0.4221 0.3178
1.1573 12.77 10800 0.4098 0.3084
1.1416 13.24 11200 0.4086 0.3119
1.1174 13.71 11600 0.3854 0.2910
1.1048 14.18 12000 0.3859 0.2824
1.0748 14.66 12400 0.3854 0.2757
1.0697 15.13 12800 0.3740 0.2724
1.0477 15.6 13200 0.3693 0.2643
1.0356 16.08 13600 0.3727 0.2561
1.0083 16.55 14000 0.3652 0.2501
1.0 17.02 14400 0.3641 0.2457
0.9779 17.49 14800 0.3568 0.2409
0.9596 17.97 15200 0.3558 0.2376
0.946 18.44 15600 0.3591 0.2311
0.9389 18.91 16000 0.3540 0.2283
0.9173 19.39 16400 0.3552 0.2265
0.9122 19.86 16800 0.3535 0.2250

Framework versions

  • Transformers 4.16.0.dev0
  • Pytorch 1.10.1+cu102
  • Datasets 1.17.1.dev0
  • Tokenizers 0.11.0