Hello! I’m new here and haven’t really used python before. My group project partner and I are making a project for class that involves creating and installation program that will work across various types of processing units provided by ACCESS (they have an AI core, a GPU core, and a CPU core.)
The proposal is to write a program that installs PyTorch 2.10.0 across all three without error and check benchmarks & such.
I was wondering if you guys could A) help me understand what exactly we’re gonna need to do in order to get it working, and B) point me to some resources I can look at to understand what’s even going on here. I’ve never used an installer or a container and don’t have a clear understanding of why this is going to be difficult enough to be a final project.
Full proposal
Concept: Installing a relatively large Python package, Pytorch, from source and also using the package through a container
Goal: Make a consistent way to install/download the python package with version consistency
Description: Pip install can vary per machine what it supplies and will change as it updates. Find out how it currently works and then create an installation process that allows a consistent download version 2.10.0 of PyTorch into a container/variable OS. Analyze the results from the ACCESSS allocations and explain why it produces the results it does
Resources: The three allocations made available by ACCESSS. Python package database. PyTorch itself.
What we plan to get from it: An understanding of the architecture of various computing systems and how to deploy installations across them.
We will be studying the PyTorch Python package, which focuses on LLM creation and running. It uses Tensors, which are the way the data is organized for the LLM, and a dynamic neural network system allowing live adjustments to information in the LLM. The intent of our research is to see how it loads in and how we can make it consistent across systems.
We will be using the ACCESSS allocations servers to track our performance metrics. We are unsure of the number of cores we will use at the current time, and we will be focusing on the execution time of the program.
We will be using Python, and don’t know which method, although it is not set in stone, as we would like to see more before finalizing our decision. For this reason, we will be starting the project with more in-depth research on the architectures and what would be good to use.