transformers
transformers copied to clipboard
Publish instance types best suited to finetune/inference of a popular model
Feature request
It would be very helpful to see a chart of instance types on the large public clouds (AWS, GCP, Oracle) most suitable for popular public LLMs like Google Flan-T5 family.
This generalizes a page from @philschmid , whose notebooks indicate how he finetuned certain models using certain instances. You could publish the instance types on the model card broken down by inference and finetuning.
Would this be possible?
Motivation
To help folks like me to not spin our wheels on trying to locate the most suitable vCPU-GPU combinations. It's a jungle on AWS for sure.
Your contribution
Happy to help how I can.