Edit model card

Whisper Medium Ro - Sarbu Vlad - multi gpu - 3

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 16.1 + Romanian speech synthesis dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0795
  • Wer: 5.5883

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
  • train_batch_size: 10
  • eval_batch_size: 10
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 3
  • total_train_batch_size: 30
  • total_eval_batch_size: 30
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 1000
  • training_steps: 10000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.1485 0.43 500 0.1439 14.2157
0.1344 0.86 1000 0.1165 11.6208
0.0745 1.3 1500 0.0930 9.7621
0.0727 1.73 2000 0.0912 9.5309
0.0269 2.16 2500 0.0802 8.1468
0.0284 2.59 3000 0.0807 8.2411
0.0162 3.03 3500 0.0765 7.6691
0.0123 3.46 4000 0.0782 7.2159
0.0199 3.89 4500 0.0794 6.9847
0.0086 4.32 5000 0.0763 6.4766
0.0083 4.75 5500 0.0768 6.6196
0.0037 5.19 6000 0.0813 6.4371
0.0035 5.62 6500 0.0780 6.0203
0.0025 6.05 7000 0.0826 6.4340
0.0032 6.48 7500 0.0763 5.7344
0.0021 6.91 8000 0.0762 5.9260
0.0011 7.35 8500 0.0790 5.5914
0.001 7.78 9000 0.0788 5.5245
0.0004 8.21 9500 0.0785 5.5883
0.0004 8.64 10000 0.0795 5.5883

Framework versions

  • Transformers 4.37.2
  • Pytorch 2.2.0
  • Datasets 2.17.0
  • Tokenizers 0.15.1
Downloads last month
2
Safetensors
Model size
817M params
Tensor type
FP16
·
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 VladS159/Whisper_medium_ro_VladS_10000_steps_multi_gpu_07_03_2024

Finetuned
(441)
this model

Dataset used to train VladS159/Whisper_medium_ro_VladS_10000_steps_multi_gpu_07_03_2024

Evaluation results

  • Wer on Common Voice 16.1 + Romanian speech synthesis
    self-reported
    5.588