ESEC/FSE 2024 Artifact Repository

Artifact repository for paper "Naming Conventions and Defects in Pre-trained Models"

About

This repository contains all of the elements of the paper.

Repo-view

The next table describes the "table of contents" for this repository.

PTM Collection ($5)

• hf_data_analyzer:

  • The scripts to anlayzed the PTM data from HuggingFace repository.

• model_collection ($5.2.1):

  • The scripts used to collect the PTM names, downloads, and the metadata from HuggingFace repository.

• converters ($5.2.2):

  • The scripts used to convert the gated models from Keras, TensorFlow, and PyTorch to the ONNX format.
  • The tests used to verify the converters.

DNN ARchitecture Assessment (DARA) ($7)

• vectorizer($7.2):

  • This folder contains the scripts used to vectorize the models.

• model_clustering ($7.3):

  • This folder contains the scripts used to cluster the models.

Name Analysis ($6, $8)

• pattern_analysis ($6):

  • This folder contains the scripts used to analyze the semantic ($6.2.1) and syntactic ($6.2.2) naming patterns.

• defect_analysis ($8):

  • This folder contains the scripts used to analyze the defects.

• manual_labeling ($8):

  • Evaluation of GPT-4 output ($6.1)
  • Manual labeling of outliers ($8.2)