Why doesn't Trisquel forum use Libre ChatGPT ?
- Login o registrati per inviare commenti
Why doesn't Trisquel forum use Libre ChatGPT ?
Hi everyone. I wanted to ask a friendly question, Why Trisquel forum does not use Libre ChatGPT?
By Libre ChatGPT I mean one of the ethical models, maybe it is LlamaGPT https://github.com/getumbrel/llama-gpt or some other. Please write which one you recommend to use !
When Avron asked about the problem with Trisquel installation I thought that it would be great if there was such a magic button somewhere at the top of the forum)))))
It's no secret that the forum is a kind of big data library.
Thank you, sam-d16, for the information about LlamaGPT.
https://trisquel.info/en/forum/chatgpt-sass
Also has information about chatgtp.
I do not know how well the
Copyright © 2004–2024 The Trisquel Project | Otherwise noted, the contents of this page are licensed under the GNU Free Documentation License with no invariant sections.
data on the forum would be the best to train machine learning training weights.
Like weights darknet uses.
I think gniall uses a dictionary like thing.
And libchatbot-eliza-perl I think is mostly a program.
I think things licensed under a free license mixed with non-free things may cause a problem with sharing training weights, but maybe large training weights for darknet or the type of file gniall uses "trained" from "ground up" could help make freely licensed functional data for machine learning programs.
There is also license compatibility to look out for.
https://www.gnu.org/licenses/license-compatibility.html
https://pjreddie.com/darknet/install/
and/or
https://trisquel.info/en/forum/darknet-floss-program
and/or
https://github.com/pjreddie/darknet.git
has more information about darknet, but in the src folder there is a cuda.c and a cuda.h file in it.
At least last time I checked.
If the input for data is non-free than the output derivative may also be non-free.
I have tried
darknet
libchatbot-eliza-perl
gniall
though there are also other chat programs in Trisquel's repository, I think.
libchatbot-eliza-perl
the text in the Synaptic Package Manager shows it is a mock Rogerian psychotherapist.
gniall attempts to "learn" whatever language you try to teach it, I think.
And darknet can "learn" how to do many things.
It is more easy for me just to do things myself than to "train" a program to do it, though maybe many people making and sharing free darknet or gniall things could make training machine learning more easy.
You may have to reprogram libchatbot-eliza-perl to make it show better text fast.
gniall may be trained faster, but may need a large amount of data to "talk" well.
darknet may work for more than just chat, but may be harder to train.
I did not know about LlamaGPT or check yet if all it's source code and data is freely licensed or if any compatibility problems are in any of it's code/art/data/other_things, but thank you for showing the link.
I do not think a chat program may work well to answer some types of questions, as it would need to be trained with the answer.
And if the answer was in the forum than a link to the answer could be shared instead of another answer in another forum post.
https://github.com/KoboldAI/KoboldAI-Client
is under the GNU AFFERO GENERAL PUBLIC LICENSE Version 3
maybe later also, but I did not test this one or check if the software has any problems.
So I do not know how well it may work, or if there is any spyware/other_problems in it.
I also do not know how this one works, or how it is "trained".
I see both KoboldAI-Client and llama-gpt have cuda things in them
https://en.wikipedia.org/wiki/CUDA
so cuda being a proprietary software/language I do not know how things like KoboldAI-Client, llama-gpt, and darknet may "legally" interact with cuda.
Mostly with KoboldAI-Client being under the GNU AFFERO GENERAL PUBLIC LICENSE Version 3 but also having a "docker-cuda" folder.
Maybe something in the GNU AFFERO GENERAL PUBLIC LICENSE Version 3 lets that be done.
Or the programmers of KoboldAI-Client can make any exception they wish.
I do not know how courts make "judicial rulings" as to show the difference between things like Linux has/uses blobs so it could break the GPL but Coreboot and Libreboot has/uses blobs but may not break the GPL.
I can only see the text of things like the GNU General Public License (GNU GPL)
but I do not know how that license gets used by people.
https://www.gnu.org/licenses/gpl-faq.html#MereAggregation
shows in part
Where's the line between two separate programs, and one program with two parts? This is a legal question, which ultimately judges will decide. We believe that a proper criterion depends both on the mechanism of communication (exec, pipes, rpc, function calls within a shared address space, etc.) and the semantics of the communication (what kinds of information are interchanged).
But the
https://sfconservancy.org/copyleft-compliance/vizio.html
shows information about an "extra element" likely, I think as I'm not a laywer, contract "law" or things like it.
So I think contracts/the_gpl are made in ways that should show how things should or should not be done, and not that each program should be decided based on just the day/case by day/case thoughts of each person/case.
Though there are still not "open and shut" cases as to what can be connected to what, and how what can interact with what.
Also how a program works on data may also sometimes be separate from a program, like taking a picture with free software and editing it with free software.
The programs are free, but both the input picture and edited picture may not be.
But machine learning may include parts of itself, as to make a derivative of the sofware included in the output picture or whatever else the AI makes.
So that may be another non "open and shut" case.
As also what for and where computers and robots are used for or at may change at some jurisdictions.
Some machine learning could be helpful, like a sawmill can be helpful for cutting wood or antilock breaks in cars, or other automation.
Or playgrounds also having recreational equipment mostly for children.
But even though
https://www.gnu.org/philosophy/free-sw.html#four-freedoms
shows in part
The freedom to run the program as you wish, for any purpose (freedom 0).
The freedom to study how the program works, and change it so it does your computing as you wish (freedom 1). Access to the source code is a precondition for this.
The freedom to redistribute copies so you can help others (freedom 2).
The freedom to distribute copies of your modified versions to others (freedom 3). By doing this you can give the whole community a chance to benefit from your changes. Access to the source code is a precondition for this.
care should also be taken when training/using automation/machine_learning so as not to break the law.
I.E. things like playgrounds and sawmills may not mix well, for legal/moral/ethical purposes instead of just license purposes.
https://www.gnu.org/licenses/license-compatibility.html
Hopefully future court cases can be more "open and shut" or dismissed in favor of the GPL/freesoftware/free_cultural_works.
That way people can have as clear an understanding of what can and can not be done with things under the GPL/free_licenses, as much as people know difference between a things like a sawmill and a playground.
I am a translator!
I have never used any large language model (LLM) myself, I only have read reports from several people who did. For example https://www.bortzmeyer.org/onprem-debut.html (he tried running them with free software only). He also posted a number of interesting experiences on the Fediverse that I am not sure it is possible to find again.
From what I understand, LLMs provide what looks most like an answer based on texts used for training. This is a kind of text inference. If asking something that requires no logic and for which there is correct text available, the LLM usually provides something accurate. Otherwise, it may be not accurate at all and no doubt is expressed, so it is highly misleading.
Another aspect reported by people who tried running such models: this takes a huge processing power. Training is even far worse.
This makes me think that the gain vs. cost is perhaps not worth it.
Hello Avron.
Thank you for your opinion. All your arguments are correct and I completely agree with you!
I thought about "Trisquel-LLM", that is, to choose a model that respects the freedom of users and launch it here and train it, and also use the answers that are already on the forum. And a living person will control the answers. Live communication is a very important element of the forum, I would even say primary, but LLM is the reality of today, it is developing very quickly and not to use this tool in my opinion would be wrong, maybe I'm wrong...
Darknet, at least the one Trisquel uses and/or the one at https://github.com/pjreddie/darknet.git
show in the LICENSE file
YOLO LICENSE
Version 2, July 29 2016
THIS SOFTWARE LICENSE IS PROVIDED "ALL CAPS" SO THAT YOU KNOW IT IS SUPER
SERIOUS AND YOU DON'T MESS AROUND WITH COPYRIGHT LAW BECAUSE YOU WILL GET IN
TROUBLE HERE ARE SOME OTHER BUZZWORDS COMMONLY IN THESE THINGS WARRANTIES
LIABILITY CONTRACT TORT LIABLE CLAIMS RESTRICTION MERCHANTABILITY. NOW HERE'S
THE REAL LICENSE:
0. Darknet is public domain.
1. Do whatever you want with it.
2. Stop emailing me about it!
Public domain is not a license, but I see what Joseph Redmon may be doing about Darknet code and data.
Unless the Stop emailing me about it! had anything to do with duress in it.
It also has
MIT License
Copyright (c) 2017 Joseph Redmon
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
the GNU GENERAL PUBLIC LICENSE Version 3, maybe later also
the
YOLO LICENSE
Version 1, July 10 2015
THIS SOFTWARE LICENSE IS PROVIDED "ALL CAPS" SO THAT YOU KNOW IT IS SUPER
SERIOUS AND YOU DON'T MESS AROUND WITH COPYRIGHT LAW BECAUSE YOU WILL GET IN
TROUBLE HERE ARE SOME OTHER BUZZWORDS COMMONLY IN THESE THINGS WARRANTIES
LIABILITY CONTRACT TORT LIABLE CLAIMS RESTRICTION MERCHANTABILITY SUBJECT TO
THE FOLLOWING CONDITIONS:
1. #yolo
2. #swag
3. #blazeit
the (I'm changing parts of this one to edit out cussing)
DO WHAT THE Edited_out_cussing YOU WANT TO PUBLIC LICENSE
Version 2, December 2004
Copyright (C) 2004 Sam Hocevar <name at domain>
Everyone is permitted to copy and distribute verbatim or modified
copies of this license document, and changing it is allowed as long
as the name is changed.
DO WHAT THE Edited_out_cussing YOU WANT TO PUBLIC LICENSE
TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
0. You just DO WHAT THE Edited_out_cussing YOU WANT TO.
and the
META-LICENSE
Version 1, June 21 2017
Any and all licenses may be applied to the software either individually
or in concert. Any issues, ambiguities, paradoxes, or metaphysical quandries
arising from this combination should be discussed with a local faith leader,
hermit, or guru. The Oxford comma shall be used.
so the GNU Free Documentation License with no invariant sections of GFDL v1.3
https://www.gnu.org/licenses/fdl-1.3-faq.html
that Trisquel uses in the forum is likely legally compatible with Darknet, I think.
I do not remember if attribution for each forum post will also be needed in the training weights that are used in Darknet.
But the Darknet program also shows text in part of
Darknet is an open source neural network framework written in C and CUDA.
So first the non-free CUDA parts may have to be removed, or Trisquel may have to remove Darknet or be de-listed as a free distro.
I do not know if Trisquel has removed the CUDA parts yet.
https://packages.trisquel.org/aramo/darknet
still shows information about non-free CUDA being part of Darknet.
I did not yet check legal compatibility of the other programs.
The CUDA parts may be under a proprietary software license in Darknet, as CUDA is proprietary.
https://en.wikipedia.org/wiki/CUDA
Though if CUDA is removed this may have compatibility with most things, and not make it's training weights proprietary, I think.
Legal disclaimer, I'm not a lawyer.
I did not check if Darknet works without CUDA's proprietary_files/information/building_parts in Trisquel's source of the program CUDA, but these files likely could be removed from Trisquel's packages repository to comply with the
Free System Distribution Guidelines (GNU FSDG)
https://www.gnu.org/distros/free-system-distribution-guidelines.html
as this shows in part
A free system distribution must not steer users towards obtaining any nonfree information for practical use, or encourage them to do so. The system should have no repositories for nonfree software and no specific recipes for installation of particular nonfree programs. Nor should the distribution refer to third-party repositories that are not committed to only including free software; even if they only have free software today, that may not be true tomorrow. Programs in the system should not suggest installing nonfree plugins, documentation, and so on.
So that is why I was typing about the non-free "optionally depends" of CUDA in Trisquel's repository for Darknet's program's source, so users can avoid them till these are removed from being in Trisquel's repository.
https://trisquel.info/en/forum/darknet-floss-program
Shows more information about where the CUDA files may be at in Trisquel's source of Darknet.