Might consider attribution
@FormerlyChuckSneedenberg
Being mean and annoying will not help us here.
@arthurmensch
I think open sourcing things would be beneficial for you and your company.
With mixtral and mistral 7b you've got quite a good reputation. I do not want another "Open"AI situation.
So did he avoid the watermarking by quantizing it?
This is the funniest pull request of all time.
So did he avoid the watermarking by quantizing it?
Mistral shared a quant version of it with some partners, that's what this is. Possible the lower quants are re-quantized though.
any chance you'd drop the whole thing now that the cat is out of the bag?
Bro probably got booted lol
also if you guys did not see this:
Thank you for being so cool with this, looking forward to the future of Mistral!
Plan: trusted
Doubters: BTFO
F16: confirmed?!
Works on my machine. Looks good to merge. ππ»
woah!
Here to request dropping full fp16 weights, if possible! π
@arthurmensch
I should also note that it was converted back to fp16 multiple times now, so it would make no sense to not release it entirely.
I know you are a company and need to ear money, but please try other ways than monitizing the model itself, even donations can help.
We as a community would be really thankful if you could release this model officially.
Weights converted to fp16 and the original fp16 weights are obviously not the same (the latter are better).
Based.
@miqudev
you should post fp16/q8 (dunno what quant you have) now, you have nothing to lose!
its real
All I want to say is leaking models is uncool but I love how Mistral has handled this.
They are seeming to go the openai route which btw.. i do not like.
At that point just release fp16.
so is this leak gonna go the llama1 route of being non-commercial used exclusively by redditors for erp or is it gonna go the llama2 route
They are seeming to go the openai route which btw.. i do not like.
They have to make money somehow. I think people like to conveniently forget that.
based.
Waiting for the blokes Q2 of this cos the Q5 was too slow to be usable with 2x 3090 but Llama license should apply here imo.
I am posting in an epic thread.
we need a license, i suggest "apache-2.0-but-leaked"
lgtm!
I was here.
Dear @arthurmensch , as a French quant to a French FP16, I'm asking you ; please publish the FP16.
This alpha is showcasing very well French tech in general and Mistral in particular, while being probably obsolete from your company standpoint : that's a win-win to surf on this wave rather than just wave at it!
The folks who will use it and tune it here will be among your best ambassadors!
Allez Arthur, sois LE MENSCH, et fais pΓ©ter le fp16 !
Goat PR
we need a license, i suggest "apache-2.0-but-leaked"
Oh yes, that would be pretty funny, please make it happen !
Posting Bizmuth in an epic thread
I Propose: OUR-MIQU
OUR MODEL
This pull request will go down in history lmao
LGTM
F16 plz!
Heh.. so miqudev probably only has GGUF and no F16 to post.
I was here too
Hi I heard there was free coffee here.
This is crazy!
Hi mom, I am on the Netflix mini series about Mistral
That's so cool, I'm sure it's historic!
Be still mine beating lungs...
Hello Netflix viewers!
Gonna need the name of that ex-employee for attribution. Would be epic if Mensch just goes full-sail into this and outs him in the PR.
Hi mum, I'm on TV!
Might consider attribution df370c87
You guys are hands down my favorite AI lab.
cool
This will surely be in a documentary in the future
Cant wait to see what else you guys have in petto. Especially as you stated that this is an old version of your flagship and you improved it by a lot!
Open source with anime icon go brrrrrrr
first
How come I am feeling like I missed something important here? is it Bizmuth (gesture based chaotic signal router) or Bismuth the metal?
Regardless, I'm sure bunch of us readily will continue using the API.
toasting in epic bread
LGTM π
Thank you for this model. It runs all the less than 2000 token queries on the CPU in under a minute on my cloud dedi instance which costs 30$ per month. Can't get the cutting edge tech cheaper than this.
This will boost Mistral initiative πππ. LGMT π
now this is epic
2024 is wild lmao.
@axra For the company yes, for me not because i want the open source community to go forward.
Mistral is already doing so much for the open source community.
The same thing that openai did in start, no?
owo?!
What an excellent marketing opportunity for Mistral if they open source this, or even release it under a non-commercial license.
betting $1 mistral gonna release fp16 medium in 7 days owo
So this model will become part of the Open Source community despite having been a leak?
Mistral should do a PR stunt and open source this just like llama and meta.
At this point, this thread has made it so hard for the people from mistral to be legally associated with this model, I doubt they will ever reply again.
This is the most amazing PR I have ever seen. This leak situation couldn't have been handled better than this. Arthur and Team Mistral, you are amazing.
lgtm
This PR is GOATed
I never thougt aout the intersection of LLM and anime. The team of Mistral is not very big, it should be easy to find the guy into Miku.
Legendary, just legendary. If anything this whole ordeal was a big PR win for Mistral.
Legendary, just legendary. If anything this whole ordeal was a big PR win for Mistral.
Absolutely! Nothingβabsolutely nothingβis bad for Mistral in all this. Everything is positive and all is good! They should continue their support and commitment to the open-source community, as they have already done.
LGTM π’
I rarely leave any kind of comments but this discussion has got me seething. @IkariDev , you and others like you are the exact reason people stop contributing to OSS. If we assume all the information on this model is accurate (which I don't see a reason for Mistral to lie about it) this is a STOLEN model. Not only that, it's not even a production model, it's an EARLY ACCESS model. So what if it's good? It might have had some deep underlying issues, it might have randomly spewed out the most vile content it can, GPT2 style. Mistral has shown nothing, but exceptional quality control, if they decided to not release this model then they had a reason to do so.
I think open sourcing things would be beneficial for you and your company.
And your qualifications are? What did you consider when coming to that conclusion?
With mixtral and mistral 7b you've got quite a good reputation. I do not want another "Open"AI situation.
Can good reputation pay the bills? Fund more compute? I think I'm safe when saying that nobody likes the business model OpenAI has adopted, but open source has been historically difficult to monetize (that's why there are new licences in the works).
I know you are a company and need to ear money, but please try other ways than monitizing the model itself, even donations can help.
I would appreciate if you could point me to a source where they stated that the model will be closed. It's an early access model, when developing this kind of technology you need to test it thoroughly. Would you like, for an example if car companies started giving anyone access to random cars they are prototyping at the moment? Also I'm really curious what "other ways" of monetization you might have in mind. You propose donations. Which OSS developer have you donated money to? Have you supported the creator of llama.cpp? Text Generation WebUI or whichever tool you use to run models? Maybe you supported Hugging Face for hosting literall terabytes of models on their site, free to download for anyone?
I should also note that it was converted back to fp16 multiple times now, so it would make no sense to not release it entirely.
yes, they are still fp16 weights of the model tho, and you can train on them ;)
If you made a game and someone post a decomp of it would you publish the entire code and assets as it "would make no sense to not release it entirely"? Hell no, you would DMCA them to oblivion and possibly sue. Mistral let this model stay and you still demand more? Listen, I don't want to mindlessly flame you, I just want you and others to realize that it's insane to demand OSS. The tone of my message is angry and that's because I am. quite. angry. after reading all this, but I just want people to think more about how they're acting. And for the love of all things please don't think that you're entitled to everything someone makes just because they contribute to OSS. If anything, they already showed great support when they didn't have to. Respect them and give back instead of only leeching off their work.
"And your qualifications are?"
None
"What did you consider when coming to that conclusion?"
Idk just my gut
"Can good reputation pay the bills?"
If you do it right yea maybe
"I would appreciate if you could point me to a source where they stated that the model will be closed"
"Would you like, for an example if car companies started giving anyone access to random cars they are prototyping at the moment?"
AI models and cars that can potentially kill someone(if published while still in prototyping phase) is a COMPLETELY different thing.
"Also I'm really curious what "other ways" of monetization you might have in mind"
i dont f know lol
"Which OSS developer have you donated money to? Have you supported the creator of llama.cpp? Text Generation WebUI or whichever tool you use to run models? Maybe you supported Hugging Face for hosting literall terabytes of models on their site, free to download for anyone?"
Im poor
"Mistral let this model stay and you still demand more?"
yes
"I just want you and others to realize that it's insane to demand OSS"
Is it really that insane when you think about f companies like OpenAI doing what they do? I just don't want this company to become OpenAI 2.0.
"The tone of my message is angry and that's because I am. quite. angry"
Don't care, my day was shit, i am angry too because companies that i like keep backstabbing me with "nuh uh no opensource for you anymore".
"And for the love of all things please don't think that you're entitled to everything someone makes just because they contribute to OSS"
I have trust issues
"Respect them and give back instead of only leeching off their work."
I respect mistral more than my right nut, its still worth a try making me look like a dumbass.. BUT getting a new open sourced base.
Also your name is dog l0ver, i can't trust you.
Anyway, have a nice day.
The tone of my message is angry and that's because I am. quite. angry.
Good. Now stfu. Cope and seethe. We won.
Thank you for this model. It runs all the less than 2000 token queries on the CPU in under a minute on my cloud dedi instance which costs 30$ per month. Can't get the cutting edge tech cheaper than this.
Could you explain which cloud dedi instance you have? I'm looking to do something similar. Thanks.
Thank you for this model. It runs all the less than 2000 token queries on the CPU in under a minute on my cloud dedi instance which costs 30$ per month. Can't get the cutting edge tech cheaper than this.
Could you explain which cloud dedi instance you have? I'm looking to do something similar. Thanks.
Sure! I am using hetzner's auctioned servers, you can find them here
@dog3-l0ver owo owo owu uwo uwu owo owo ovo
I rarely leave any kind of comments but this discussion has got me seething. @IkariDev , you and others like you are the exact reason people stop contributing to OSS. If we assume all the information on this model is accurate (which I don't see a reason for Mistral to lie about it) this is a STOLEN model. Not only that, it's not even a production model, it's an EARLY ACCESS model. So what if it's good? It might have had some deep underlying issues, it might have randomly spewed out the most vile content it can, GPT2 style. Mistral has shown nothing, but exceptional quality control, if they decided to not release this model then they had a reason to do so.
I think open sourcing things would be beneficial for you and your company.
And your qualifications are? What did you consider when coming to that conclusion?
With mixtral and mistral 7b you've got quite a good reputation. I do not want another "Open"AI situation.
Can good reputation pay the bills? Fund more compute? I think I'm safe when saying that nobody likes the business model OpenAI has adopted, but open source has been historically difficult to monetize (that's why there are new licences in the works).
I know you are a company and need to ear money, but please try other ways than monitizing the model itself, even donations can help.
I would appreciate if you could point me to a source where they stated that the model will be closed. It's an early access model, when developing this kind of technology you need to test it thoroughly. Would you like, for an example if car companies started giving anyone access to random cars they are prototyping at the moment? Also I'm really curious what "other ways" of monetization you might have in mind. You propose donations. Which OSS developer have you donated money to? Have you supported the creator of llama.cpp? Text Generation WebUI or whichever tool you use to run models? Maybe you supported Hugging Face for hosting literall terabytes of models on their site, free to download for anyone?
I should also note that it was converted back to fp16 multiple times now, so it would make no sense to not release it entirely.
yes, they are still fp16 weights of the model tho, and you can train on them ;)If you made a game and someone post a decomp of it would you publish the entire code and assets as it "would make no sense to not release it entirely"? Hell no, you would DMCA them to oblivion and possibly sue. Mistral let this model stay and you still demand more? Listen, I don't want to mindlessly flame you, I just want you and others to realize that it's insane to demand OSS. The tone of my message is angry and that's because I am. quite. angry. after reading all this, but I just want people to think more about how they're acting. And for the love of all things please don't think that you're entitled to everything someone makes just because they contribute to OSS. If anything, they already showed great support when they didn't have to. Respect them and give back instead of only leeching off their work.
I think you need a big hug. Also tend to your dog.
Insane, that's all I've to say.
I love Mistral
miqu won!
we did it reddit!
we did it reddit!
Miqu keeps winning
I rarely leave any kind of comments but this discussion has got me seething. @IkariDev , you and others like you are the exact reason people stop contributing to OSS. If we assume all the information on this model is accurate (which I don't see a reason for Mistral to lie about it) this is a STOLEN model. Not only that, it's not even a production model, it's an EARLY ACCESS model. So what if it's good? It might have had some deep underlying issues, it might have randomly spewed out the most vile content it can, GPT2 style. Mistral has shown nothing, but exceptional quality control, if they decided to not release this model then they had a reason to do so.
I think open sourcing things would be beneficial for you and your company.
And your qualifications are? What did you consider when coming to that conclusion?
With mixtral and mistral 7b you've got quite a good reputation. I do not want another "Open"AI situation.
Can good reputation pay the bills? Fund more compute? I think I'm safe when saying that nobody likes the business model OpenAI has adopted, but open source has been historically difficult to monetize (that's why there are new licences in the works).
I know you are a company and need to ear money, but please try other ways than monitizing the model itself, even donations can help.
I would appreciate if you could point me to a source where they stated that the model will be closed. It's an early access model, when developing this kind of technology you need to test it thoroughly. Would you like, for an example if car companies started giving anyone access to random cars they are prototyping at the moment? Also I'm really curious what "other ways" of monetization you might have in mind. You propose donations. Which OSS developer have you donated money to? Have you supported the creator of llama.cpp? Text Generation WebUI or whichever tool you use to run models? Maybe you supported Hugging Face for hosting literall terabytes of models on their site, free to download for anyone?
I should also note that it was converted back to fp16 multiple times now, so it would make no sense to not release it entirely.
yes, they are still fp16 weights of the model tho, and you can train on them ;)If you made a game and someone post a decomp of it would you publish the entire code and assets as it "would make no sense to not release it entirely"? Hell no, you would DMCA them to oblivion and possibly sue. Mistral let this model stay and you still demand more? Listen, I don't want to mindlessly flame you, I just want you and others to realize that it's insane to demand OSS. The tone of my message is angry and that's because I am. quite. angry. after reading all this, but I just want people to think more about how they're acting. And for the love of all things please don't think that you're entitled to everything someone makes just because they contribute to OSS. If anything, they already showed great support when they didn't have to. Respect them and give back instead of only leeching off their work.
I think you need a big hug. Also tend to your dog.
as an early ml engineer, i find it really hard to make money from donations, they are doing the right thing, specially as they are a company, a company just doesn't survive without money
I'm glad to be part of the history, you guys rock!
First torrent release and now this haha can someone dream to be part of Mistral Team!
Hope the best for Mistral team!
Keep the amazing work.