Skip to main content

HF Report

Created on October 20|Last edited on October 20

Section 1

My findings.
/


Weights & Biases

Text Classification on the "Banking77" dataset

This W&B Workspace was created using AWS SageMaker and includes:

  • hyperparameter tuning carried out over the model_name, learning_rate and warmup_steps hyperparameters. You can filter this workspace by filtering the Job Type column to "HyperparameterTuning" to see only these experiment runs
  • W&B Tables for exploratory data analysis of the raw dataset
  • A W&B Report exploring the dataset and exploring the results of these experiments

See here for the full SageMaker & Hugging Face training scripts and our docs for futher into

Run set
27



<null>albert-large-v2distilbert-base-uncasedgoogle/electra-large-discriminatorroberta-largemodel_name_or_pathnull020406080100120140160180200220240260warmup_stepsnull0.000000.000050.000100.000150.000200.000250.000300.000350.00040learning_ratenull0.00.51.01.52.02.53.03.54.04.5eval/lossnull0.00.10.20.30.40.50.60.70.80.91.0eval/accuracy
Run set
27