Proposal: Pytorch NFT checkpoint

Name of Project: Pytorch NFT checkpoint

Proposal in one sentence: A custom pytorch NFT-checkpoint that hashes the current network weights, some metadata (data, accuracy, etc…) and your eth address (could be turned into some kind of standard later) every N epoch, which proves who did the network training.

This NFT could represent a tradable license of some sort.

Description of the project and what problem is it solving: Trying to determine who actually did the training of a machine learning model is currently very hard. Looking at open-source model zoos today it is nearly impossible to determine who trained which model. The Pytorch NFT checkpoint solves this problem.

A custom NFT is generated each epoch, which proves who generated which network weights. This could be used as a badge of honor or turned into a tradable license.

Grant Deliverables:

  • Working Pytorch NFT checkpoint implementation
  • Easily installable for the latest Pytorch version

Funding Requested: $1000




Hey, it’s a cool project proposal. The hashed weights that you want to store as NFTs brings up similarities between my proposal: Quantifying Trust for prescribing drugs wherein I have weights hashed and shared over a recombination engine for doctors when they’re writing drugs. The shared weights will change and tune as more prescriptions are written, i.e its uniqueness becomes a function of time. We might be able to find some common ground on both projects and inter-use a few components.

Thats super interesting!

Add me on discord (shafu.eth#4727), would love to share some ideas I have with you.