mitdbg / palimpzest

A Declarative System for Optimizing AI Workloads
https://dsg.csail.mit.edu/projects/palimpzest/
MIT License
46 stars 5 forks source link

Error with pz.Execute() when installing from source #22

Open aychang95 opened 4 months ago

aychang95 commented 4 months ago

I installed palimpzest from source and tried to run the example notebook but the line below gave an error that Execute() method does not exist:

results, plan = pz.Execute(emails, policy, num_samples=2, nocache=True, verbose=True)

Works well when installing via PyPI: pip install palimpzest.

mdr223 commented 4 months ago

Hi @aychang95, thanks for opening an issue!

You're absolutely correct that the Execute() functionality will error at the moment if you install from source. Shortly after releasing the paper pre-print we began embarking on an overdue refactor of our code, which is slated to be merged back into main on June 17th.

If the pip package is not sufficient for your use case, please let me know and I'd be happy to help debug things so you can get a working version in the 5-day interim.

Also, if you don't mind my asking, what is your intended use case for PZ? We're very interested in learning more about the problems / workloads people are trying to solve with this system.

aychang95 commented 4 months ago

Ah I saw prof. @mikecafarella speak on Palimpzest at the MIT CSAIL Summit event last week, and found the premise super interesting and potentially useful for our python notebook environments. If you're almost done with the refactoring, I can definitely wait to try it out.

I've personally been wanting to experiment it from source with our GPU notebook servers. I'm curious to see how it could work with locally running LLMs.

For now we've got the pip package working in our palimpzest notebook environment on our CPU and GPU servers your demo as a quickstart notebook:

Open In American Data Science

Also as an aside, I would love the ability to set the gradio port param as well e.g. 5000 (instead of the default)