search

SakanaAI / evolutionary-model-merge

Official repository of Evolutionary Optimization of Model Merging Recipes
Apache License 2.0
1.08k stars 72 forks source link

This is literally garbage. you made garbage #2

Open rombodawg opened 3 months ago

rombodawg commented 3 months ago

Releasing a github page meant to share a new merging method, only to provide no merging code whatsoever is a really shitty move. Its not so much the fact that you didnt release your code that sucks. Its that you acted like you were going to, and put up a front like you are going to open source your methods, and then clearly only did it for popularity. Just a waste of an hour I spend researching this. Good job. 👎👎

rombodawg commented 3 months ago

@iwiwi @lerrytang @hardmaru @mkshing ^

Im assuming these are the researchers on the team at SakanaAi, since those are the only names that came up when I @ by default

thistleknot commented 3 months ago

are you saying the code on the repo isn't doing what the readme says and/or doesn't meet the paper's expectations? I was interested in this as well, but when I ran it, was told my gpu wasn't up to the compute requirements.

oliveirabruno01 commented 3 months ago

@thistleknot the code does exactly what the README says. Unfortunately this is just the evaluation code, to evaluate the models they created if you want. But they do say this:

Look forward to more updates and additions coming soon.

So let's hope that this includes the evolutionary algorithms they used. It's a great project and there's no reason to anyone to dunk on the authors. But we CAN watch this repo and hope that they launch the research code, or someone else recreates it.

If that's what you're looking for I suggest you to "Watch" this repo rather than give it a "Star" just yet, and to upvote the first comment on #3 so the authors see that there's interested people out there.

In the meanwhile, there's some people trying to implement their discoveries here on github:

Though an Official implementation would be very appreciated.

JinSeoung-Oh commented 3 months ago

I somewhat agree with what you're saying. I was really excited when I found this GitHub after reading the paper and eagerly looked through the code. However, I was quite disappointed to find that there's no section on model fusion methods in this GitHub. Just to be sure, I checked the issue page... and indeed, it seems that the corresponding code hasn't been released yet. Nevertheless, since they mentioned they'll be updating it in the future, I guess we'll have to look forward to that. I just hope the owner of this GitHub doesn't pull a disappearing act like in some other cases where they promise to release the code and then vanish forever