Note
August 21, 2024 — This is brand new and there will be lots of updates coming soon.
I made this silly little tool because I'm tired of wondering, subjectively, if one prompt works better than another one. I want a way to test them against eah other. So that's what this is.
- First install Fabric, which is another project of ours. To install Fabric, make sure Go is installed, and then run the following command.
# Install Fabric directly from the repo
go install github.com/danielmiessler/fabric@latest
# Run the setup to set up your directories and keys
fabric --setup
If everything works you are good to go, but you may need to set some environment variables in your ~/.bashrc
or ~/.zshrc
file. Here is an example of what you can add:
# Golang environment variables
export GOROOT=/usr/local/go
export GOPATH=$HOME/go
export PATH=$GOPATH/bin:$GOROOT/bin:$HOME/.local/bin:$PATH:
- Then once
fabric
runs fine, you're pretty much done.harness
just runs Fabric usingBash
.
To use harness
, just do the following:
- Clone this repo.
cd $your_harness_directory
- Put your input in
input.md
(like the transcript you're analyzing, or whatever) - Put your first prompt in
prompt1.md
. - Put your second prompt in
prompt2.md
. - Run
./harness.sh
.
Harness has a cool feature where you can try mulitple runs and do analysis on the full set of results (because LLMs have a lot of various from run to run).
To do that, just add a number to the end of the command.
./harness.sh 10
Now you should be able to see which is better across multiple runs, and if you don't see much of a difference across like 10 runs, the differences are probably pretty small.
Enjoy!
Caution
There is no security or input validation of any kind on this thing. It's a shell script, so like don't put this in production or anything.