Edit model card

hello_2b_2

This model is a fine-tuned version of facebook/wav2vec2-xls-r-2b on the COMMON_VOICE - TR dataset. It achieves the following results on the evaluation set:

  • Loss: 0.5324
  • Wer: 0.5109

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: 5e-06
  • train_batch_size: 2
  • eval_batch_size: 8
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 2
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 32
  • total_eval_batch_size: 16
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 30.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
3.3543 0.92 100 3.4342 1.0
3.0521 1.85 200 3.1243 1.0
1.4905 2.77 300 1.1760 0.9876
0.5852 3.7 400 0.7678 0.7405
0.4442 4.63 500 0.7637 0.7179
0.3816 5.55 600 0.7114 0.6726
0.2923 6.48 700 0.7109 0.6837
0.2771 7.4 800 0.6800 0.6530
0.1643 8.33 900 0.6031 0.6089
0.2931 9.26 1000 0.6467 0.6308
0.1495 10.18 1100 0.6042 0.6085
0.2093 11.11 1200 0.5850 0.5889
0.1329 12.04 1300 0.5557 0.5567
0.1005 12.96 1400 0.5964 0.5814
0.2162 13.88 1500 0.5692 0.5626
0.0923 14.81 1600 0.5508 0.5462
0.075 15.74 1700 0.5477 0.5307
0.2029 16.66 1800 0.5501 0.5300
0.0985 17.59 1900 0.5350 0.5303
0.1674 18.51 2000 0.5429 0.5241
0.1305 19.44 2100 0.5645 0.5443
0.0774 20.37 2200 0.5313 0.5216
0.1372 21.29 2300 0.5644 0.5392
0.1095 22.22 2400 0.5577 0.5306
0.0958 23.15 2500 0.5461 0.5273
0.0544 24.07 2600 0.5290 0.5055
0.0579 24.99 2700 0.5295 0.5150
0.1213 25.92 2800 0.5311 0.5221
0.0691 26.85 2900 0.5228 0.5095
0.1729 27.77 3000 0.5340 0.5095
0.0697 28.7 3100 0.5334 0.5139
0.0734 29.63 3200 0.5323 0.5140

Framework versions

  • Transformers 4.13.0.dev0
  • Pytorch 1.10.0
  • Datasets 1.15.2.dev0
  • Tokenizers 0.10.3
Downloads last month
3

Dataset used to train patrickvonplaten/hello_2b_2