CoolFace
Modelpublic

RobbieJimersonJr/whisper-medium-Bambara-field

sourceHugging Faceapache-2.0updated 4y agoView on Hugging Face
3likes46downloads
Model Card

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

Whisper Medium Bambara Fieldwork

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

  • —Loss: 4.0725
  • —Wer: 157.6004

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: 1e-05
  • —trainbatchsize: 16
  • —evalbatchsize: 2
  • —seed: 42
  • —distributed_type: multi-GPU
  • —optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • —lrschedulertype: linear
  • —lrschedulerwarmup_steps: 500
  • —training_steps: 13532
  • —mixedprecisiontraining: Native AMP

Training results

Training LossEpochStepValidation LossWer
0.63421.0310002.5810159.5293
0.28733.0320002.9513159.1140
0.14615.0230003.6833158.3941
0.0497.0240004.0725157.6004
0.02189.0150004.2531158.3664
0.007111.060004.5944157.9972
0.005712.0470004.6659161.4952
0.006114.0380004.9162161.0614
0.004216.0390005.0205158.9848
0.000718.02100005.1463159.1970
0.001520.01110005.2150159.0401
0.000122.01120005.3008159.5478
0.000124.0130005.3800159.0309

Framework versions

  • —Transformers 4.25.1
  • —Pytorch 1.13.1+cu117
  • —Datasets 2.8.0
  • —Tokenizers 0.13.2