CoolFace
Modelpublic

NbAiLabArchive/scream_tertius_dropout_replicate_test7a

sourceHugging Faceapache-2.0updated 3y agoView on Hugging Face
0likes33downloads
Model Card

<!-- This model card has been generated automatically according to the information Keras had access to. You should probably proofread and complete it, then remove this comment. -->

screamtertiusdropoutreplicatetest7a

This model is a fine-tuned version of openai/whisper-small on the NbAiLab/NCCspeechall_v5 dataset. It achieves the following results on the evaluation set:

  • step: 19999
  • eval_loss: 1.1799
  • train_loss: 0.3145
  • eval_wer: 11.6626
  • eval_cer: 5.8907

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: 2e-05
  • lrschedulertype: linear
  • perdevicetrainbatchsize: 32
  • totaltrainbatchsizeper_node: 128
  • totaltrainbatch_size: 1024
  • totaloptimizationsteps: 20,000
  • startingoptimizationstep: None
  • finishingoptimizationstep: 20,000
  • numtraindataset_workers: 32
  • num_hosts: 8
  • totalnumtraining_examples: 20,480,000
  • stepsperepoch: 1314
  • num_beams: 5
  • dropout: True
  • dropout_probability: 0.1

Training results

stepeval_losstrain_losseval_wereval_cer
01.35827.9231169.1230127.5435
10000.92030.974824.02569.2618
20000.99510.674718.75767.4326
30001.10730.549516.74797.1000
40001.10930.461214.43366.4147
50001.17190.432614.19006.2837
60001.26270.399812.81975.9814
70001.27850.376512.78936.1476
80001.13950.386912.51526.0519
90001.23270.361612.78936.1829
100001.08550.362011.44955.6790
110001.10180.345311.75405.7848
120000.99530.348611.72355.7294
130001.13210.336512.02806.0015
140001.26540.333511.63225.8050
150001.21490.306111.84535.8503
160001.15390.309011.93675.8503
170001.25300.310311.75405.8251
180001.19250.320911.47995.6790
190001.21550.293111.72355.9562
199991.17990.314511.66265.8907

Framework versions

  • Transformers 4.29.0.dev0
  • Datasets 2.12.0
  • Tokenizers 0.13.3