2.1 KiB
LLPTE
LLPTE (LLM Prompt Technique Experiments). Repository to experiment with different prompting techniques for translating Gherkin Acceptance Criteria into Rimay (CNL) System Requirements
Setup
Use conda and enable extra channels (such as conda-forge)
Deactive conda:
conda deactivate
conda env remove -n smell-detector
Active conda:
conda env create -f conda_llpte.yml
conda activate smell-detector
Run this application with python:
- Ubuntu version used: 22.04, full desktop edition. (server is also fine)
- Install requirements:
pip install -r requirements.txt
python3 main.py
Make sure to open port 8080 on webserver to access nicegui For standalone modes add the parameter: standalone
Results will be logged in prompt_logging folder, sorted by LLM-prompting techniqnue. Filename contains score from DSL Rimay and Paska.
python3 main.py standalone 0 20
Means run script standalone and use acceptance criteria 0 - 20.
Input dataset for Gherkin is inspired by https://zenodo.org/records/6460854 See input_dataset.xml
Notes to run paska tooling
use full Java 8 JDK with JFX runtime. Do not use the small version of java on Linux Ubuntu. I also changed the conda environment file to be compatible with Ubuntu instead of OSX.
To install correct version of JDK: https://bell-sw.com/pages/repositories/#apt
wget -q -O - https://download.bell-sw.com/pki/GPG-KEY-bellsoft | sudo apt-key add -
echo "deb [arch=amd64] https://apt.bell-sw.com/ stable main" | sudo tee /etc/apt/sources.list.d/bellsoft.list
sudo apt-get update
sudo apt-get install bellsoft-java8-runtime-full
Also download the following distsim tagger file:
https://github.com/amiangshu/SentiSE/blob/master/edu/stanford/nlp/models/pos-tagger/english-left3words/english-left3words-distsim.tagger
And see the following file for needed modifications to paths: /LLPTE/rimay_verification.py
Also, there is no error when the directory of preproccesing tooling is not found, be aware. And use full paths for everything.
I precompiled the jar files for Paska and DSL Rimay.