7 ms·

Good point - the main issue is we encountered this exact issue with our old package Hyperlearn (https://github.com/danielhanchen/hyperlearn https://github.com/d

Good point - the main issue is we encountered this exact issue with our old package Hyperlearn (https://github.com/danielhanchen/hyperlearn https://github.com/danielhanchen/hyperlearn). I OSSed all the code to the community - I'm actually an extremely open person and I love contributing to the OSS community. The issue was the package got gobbled up by other startups and big tech companies with no credit - I didn't want any cash from it, but it stung and hurt really bad hearing other startups and companies claim it was them who made it faster, whilst it was actually my work. It hurt really bad - as an OSS person, I don't want money, but just some recognition for the work. I also used to accept and help everyone with coding up their startup's software, but I never got paid or even any thanks - sadly I didn't expect the world to be such a hostile place. So after a sad awakening, I decided with my brother instead of OSSing everything, we would first OSS something which is still very good - 5X faster training is already very reasonable. I'm all open to other suggestions on how we should approach this though! There are no evil intentions - in fact I insisted we OSS EVERYTHING even the 30x faster algos, but after a level headed discussion with my brother - we still have to pay life expenses no? If you have other ways we can go about this - I'm all ears!! We're literally making stuff up as we go along!

kristopolous 3y ago
There's a variety of licenses some of which protect you from this problem. Maybe those would be appropriate?
- danielhanchen 3y ago
  Sadly I've tried GPL3 for Hyperlearn and AGPL3 - those licenses only work for UI or Database based OSS - eg MongoDB, Gitlab etc. I still haven't figured out how to open source algorithms which don't have an UI or database - maybe a training or inference platform? But if we made a training / inference platform, then there won't be any OSS code. We're currently stuck in the middle - do you have any suggestions on this?
  - Jayakumark 3y ago
    Sell binaries without code for individuals/enterprises who can train on private data , Get some GPUs and sell training platforms similar to together.ai for enterprises with data privacy guarantee. just make it super simple , give a google like sheet to fill or dump a folder with text files and you can use ml to make it in structured format. User gives Input - > excel, txt etc Select model base -> mystal, llama2 output --> fine trained model. Give a fixed cost based on data you got .. say it will take $50 to train on this ? for 25 hours of GPU with your markup. User Pays the bill and get PEFT llama model out. The big problem i see in Finetuning for GPU Poor is no one gives an estimate on how much it will cost for your data on a given model. have a calculator for x words or Bytes and y epochs, here is the cost we are estimating.. Once you make your $x million in terms of 50x your pay for the time you invested, you can open source it, but we are greedy when it comes to money.. so opensource model at your will. In the mean time hope no one catches your technique.
    - danielhanchen 3y ago
      Ye a training platform! Ye for hobbyists - it's hard to price. But I agree some sort of platform for training could in theory work
  - Zuiii 3y ago
    > I still haven't figured out how to open source algorithms You're not supposed to be able to restrict algorithms (i.e. math), and copyright (e.g. open source licenses) is definitely not the right tool. If you want to do it anyway, You'll have better chances by abusing the patent system.
    - danielhanchen 3y ago
      Hmmm patents - doesn't that also showcase everything in the open though?
      - Zuiii 3y ago
        Yes, but that's the only way I know of to legally stop others from making money off "ideas". Copyright isn't a good tool for this as oracle recently found out.
squigz 3y ago
Thanks for your response, and for open sourcing things at all :) > The issue was the package got gobbled up by other startups and big tech companies with no credit I understand that this is frustrating, but are you open sourcing things for recognition? The fact is companies are going to use 100s of open source packages and aren't going to credit them all, or even any. At any rate, I appreciate how difficult your position must be, and wish you luck.
- danielhanchen 3y ago
  Thanks! :) I agree, but generally as a researcher / OSS even a nice citation is nice :) There are some cool people who do just that, but some do not. But thanks once again!
nikolayasdf123 3y ago
> gobbled up by other startups and big tech companies with no credit true. saw this myself too for my projects. big tech / startups would use your project internally and give you nothing back. sadly happens often.
- danielhanchen 3y ago
  :( Very sad
- MyFirstSass 3y ago
  I feel like free open source without paying is for tinkerers, startups, smaller orgs, but as soon as you become a "big company" with wealthy owners you need to either contribute or pay up. This also incentives contributing to open source, because you'd potentially get compensated if the code was used by some big corporation. Why isn't this a thing?
  - danielhanchen 3y ago
    I also wish this was the case!

kristopolous 3y ago

There's a variety of licenses some of which protect you from this problem. Maybe those would be appropriate?

danielhanchen 3y ago
Sadly I've tried GPL3 for Hyperlearn and AGPL3 - those licenses only work for UI or Database based OSS - eg MongoDB, Gitlab etc. I still haven't figured out how to open source algorithms which don't have an UI or database - maybe a training or inference platform? But if we made a training / inference platform, then there won't be any OSS code. We're currently stuck in the middle - do you have any suggestions on this?
- Jayakumark 3y ago
  Sell binaries without code for individuals/enterprises who can train on private data , Get some GPUs and sell training platforms similar to together.ai for enterprises with data privacy guarantee. just make it super simple , give a google like sheet to fill or dump a folder with text files and you can use ml to make it in structured format. User gives Input - > excel, txt etc Select model base -> mystal, llama2 output --> fine trained model. Give a fixed cost based on data you got .. say it will take $50 to train on this ? for 25 hours of GPU with your markup. User Pays the bill and get PEFT llama model out. The big problem i see in Finetuning for GPU Poor is no one gives an estimate on how much it will cost for your data on a given model. have a calculator for x words or Bytes and y epochs, here is the cost we are estimating.. Once you make your $x million in terms of 50x your pay for the time you invested, you can open source it, but we are greedy when it comes to money.. so opensource model at your will. In the mean time hope no one catches your technique.
  - danielhanchen 3y ago
    Ye a training platform! Ye for hobbyists - it's hard to price. But I agree some sort of platform for training could in theory work
- Zuiii 3y ago
  > I still haven't figured out how to open source algorithms You're not supposed to be able to restrict algorithms (i.e. math), and copyright (e.g. open source licenses) is definitely not the right tool. If you want to do it anyway, You'll have better chances by abusing the patent system.
  - danielhanchen 3y ago
    Hmmm patents - doesn't that also showcase everything in the open though?
    - Zuiii 3y ago
      Yes, but that's the only way I know of to legally stop others from making money off "ideas". Copyright isn't a good tool for this as oracle recently found out.

squigz 3y ago

Thanks for your response, and for open sourcing things at all :) > The issue was the package got gobbled up by other startups and big tech companies with no credit I understand that this is frustrating, but are you open sourcing things for recognition? The fact is companies are going to use 100s of open source packages and aren't going to credit them all, or even any. At any rate, I appreciate how difficult your position must be, and wish you luck.

danielhanchen 3y ago
Thanks! :) I agree, but generally as a researcher / OSS even a nice citation is nice :) There are some cool people who do just that, but some do not. But thanks once again!

nikolayasdf123 3y ago

> gobbled up by other startups and big tech companies with no credit true. saw this myself too for my projects. big tech / startups would use your project internally and give you nothing back. sadly happens often.

danielhanchen 3y ago
:( Very sad
MyFirstSass 3y ago
I feel like free open source without paying is for tinkerers, startups, smaller orgs, but as soon as you become a "big company" with wealthy owners you need to either contribute or pay up. This also incentives contributing to open source, because you'd potentially get compensated if the code was used by some big corporation. Why isn't this a thing?
- danielhanchen 3y ago
  I also wish this was the case!