Don't take Apple MLX too seriously, it's not going to last, here's why (serious post)
Posted by nderstand2grow@reddit | LocalLLaMA | View on Reddit | 56 comments
People get excited about Apple's new MLX library that let's you run LLMs and other GenAI models in a way that's optimized for Apple Silicon. I'm just here to warn you: The MLX library is not going to last and here's why:
Apple already has an official Github account where they list their repositories:
When I first heard about Apple MLX, I wondered: Why isn't it on Apple's repo then? MLX is currently hosted here: https://github.com/ml-explore/mlx
The repo describes it as "Machine learning research on your laptop or in a data center - by Apple", but why isn't it on Apple's official account then? For context, Apple Ferret is on Apple's account: https://github.com/apple/ml-ferret
Hypothesis:
- MLX is just a curiosity of a team of developers at Apple. They thought of the project as a way to show to the management that Apple Silicon has potential, but since the higher-ups still don't think that LLMs (and GenAI in general) add value to the company, they refused to let the devs host the repo on Apple's official Github account.
- Since when did Apple show interest in offering customizable, hackable products? Apple's entire $3T market value is due to their walled-garden business model. The MLX library is in stark contrast to Apple's modus operandi.
- I'm all for open-source and allowing users to use the software any way they want, but projects like this are just typical one-way open-source, meaning that the company just want to harness the power of open-source to get ideas, fix bugs, silently test products, etc. so that later the company can integrate the product in their ecosystem for more profit.
Still don't believe me? Look at Apple Ferret license below. They basically took Llava and Vicuna and made them more restricted:
Copyright (C) 2023 Apple Inc. All Rights Reserved.
IMPORTANT: This Apple software is supplied to you by Apple
Inc. ("Apple") in consideration of your agreement to the following
terms, and your use, installation, modification or redistribution of
this Apple software constitutes acceptance of these terms. If you do
not agree with these terms, please do not use, install, modify or
redistribute this Apple software.
In consideration of your agreement to abide by the following terms, and
subject to these terms, Apple grants you a personal, non-exclusive
license, under Apple's copyrights in this original Apple software (the
"Apple Software"), to use, reproduce, modify and redistribute the Apple
Software, with or without modifications, in source and/or binary forms;
provided that if you redistribute the Apple Software in its entirety and
without modifications, you must retain this notice and the following
text and disclaimers in all such redistributions of the Apple Software.
Neither the name, trademarks, service marks or logos of Apple Inc. may
be used to endorse or promote products derived from the Apple Software
without specific prior written permission from Apple. Except as
expressly stated in this notice, no other rights or licenses, express or
implied, are granted by Apple herein, including but not limited to any
patent rights that may be infringed by your derivative works or by other
works in which the Apple Software may be incorporated.
The Apple Software is provided by Apple on an "AS IS" basis. APPLE
MAKES NO WARRANTIES, EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION
THE IMPLIED WARRANTIES OF NON-INFRINGEMENT, MERCHANTABILITY AND FITNESS
FOR A PARTICULAR PURPOSE, REGARDING THE APPLE SOFTWARE OR ITS USE AND
OPERATION ALONE OR IN COMBINATION WITH YOUR PRODUCTS.
IN NO EVENT SHALL APPLE BE LIABLE FOR ANY SPECIAL, INDIRECT, INCIDENTAL
OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) ARISING IN ANY WAY OUT OF THE USE, REPRODUCTION,
MODIFICATION AND/OR DISTRIBUTION OF THE APPLE SOFTWARE, HOWEVER CAUSED
AND WHETHER UNDER THEORY OF CONTRACT, TORT (INCLUDING NEGLIGENCE),
STRICT LIABILITY OR OTHERWISE, EVEN IF APPLE HAS BEEN ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.
Edit: Added my comments here for the sake of completeness:



asankhs@reddit
They seem to be doing quite well have you tried mlx-optiq.com ?
Herr_Doktor_Sly@reddit
Well, that didn't age well, did it...
westsunset@reddit
Lol just came across this and had the same thought
nderstand2grow@reddit (OP)
It aged well imo! Awni, the MLX head just left Apple after months of threatening to leave.
Sufficient_Fun2386@reddit
this definitely aged like milk.
theres now a lot of models being translated to mlx Huggingface MLX Community
LM Studio just recently supports MLX runtime along side metal llama
QuickQuirk@reddit
I just found this post today, and was thinking "This really aged well" :D
OnlyOnOkasion@reddit
https://www.bloomberg.com/news/articles/2025-06-30/apple-weighs-replacing-siri-s-ai-llms-with-anthropic-claude-or-openai-chatgpt?embedded-checkout=true
simplestpanda@reddit
Literally unrelated, but ok.
kaptenbiskut@reddit
I would choose MLX over MPS all day everyday for the 10x speed.
Significant-Pay-6476@reddit
Is it really 10x speed?
csharp-agent@reddit
So how it’s going?
nderstand2grow@reddit (OP)
you tell me, has Apple officially released any models for MLX or have they used MLX in any part of their own OS? Didn’t think so.
csharp-agent@reddit
so, what framework did you use for loca run?
Im asking becase im building some part of our app, and I wanto to run local models, so now I use LamaSharp (lama.cpp wrapper) and Im explore also other options.
BeginningVictory4408@reddit
Hi, I'm looking to do a dissertation on same idea to run SLM on iphone devices for simple tasks as I feel market will move towares no UI kind of approach and it takes only one update on iOS where it can release MLX and SLM options for developers to react with each other via apps. With app intents. I think Apple will push forward on this? What do you think?
csharp-agent@reddit
💯 models will run locally. to,reduce cost for “free“ users
nderstand2grow@reddit (OP)
i simply used llama.cpp or oobabooga
cosminmarin@reddit
This proves it to be wrong: https://www.youtube.com/watch?v=A0onppIyHEg EXO uses MLX distributed with TB5 interconnect, and it's optimised for Apple Silicon.
jailbreakerAI@reddit
intresting!
recurrence@reddit
Can you comment on the current state of MLX? It appears that your hypothesis was false but perhaps I am missing some important details.
Also in RE to "The MLX library is not going to last ": What was the time horizon for this expectation?
nderstand2grow@reddit (OP)
I mean, do you still see Apple using MLX in any project? to them, MLX is a toy project. they haven't hired enough devs to work fully on it. and they don't even open the Apple NPU code to their own team (even MLX team at apple can't access ANE)!!!
PeakBrave8235@reddit
This aged well
fremenmuaddib@reddit
Maybe this aged poorly, but surely this is still a huge issue:
NOTES ON METAL BUGS by neobundy
Apple should admit the design flaws in the M processors and start working on a better architecture if it wants to stay relevant in the AI era.
recurrence@reddit
Does this bug exist on the M4 series processors?
davewolfs@reddit
Does the bug exist on M3U or M4M?
No_Box_2220@reddit
> When I first heard about Apple MLX, I wondered: Why isn't it on Apple's repo then? MLX is currently hosted here: https://github.com/ml-explore/mlx
webkit also not in the apple's repo https://github.com/WebKit/WebKit . I think it's fine to create a repo not under company name.
Txt8aker@reddit
This guy is probably from Nvidia PR trying to negate any competition lol
lollipop3653@reddit
9 month from the future and it's getting a lot of support still.
nderstand2grow@reddit (OP)
yes, except it's not featured on Apple's open source projects website
PeakBrave8235@reddit
It is
RickySpanishLives@reddit
The amount of times the OP has been shown to be wrong is funny.
Free_Ad8975@reddit
Well this aged poorly...
HeBoughtALot@reddit
I am amused at the lengths people still go to to shit on Apple.
Meanwhile Facebook is heroic for Llama. Ignore the entire generation of young people with depression and the global wave of fascist movements. (Brought to you by Meta Platforms, Inc.)
PeakBrave8235@reddit
Fuck Fuckerberg
FullSendGod@reddit
can confirm that apple will go to war with the big boys using mlx, so this is awkward...
netroxreads@reddit
This did not age well. We have MLX out and they're definitely faster which means more tokens in less time. A few LLMs have been converted to MLX for AS.
blacktothafuture@reddit
This is so funny lmao
anzzax@reddit
What is your opinion after 10 months? Serious question: with Apple’s push into AI, I'm seriously considering investing my time in building AI-enabled apps for the Apple ecosystem. It's a long-term vision (about 5 years). Next-gen assistants require low latency and access to local context, and I believe Apple is in a very good position to enable a breakthrough in UX. I understand that local devices will never fully compete with big cloud models, but I believe in complex hybrid apps where local models handle multi-modal preprocessing, run or use tools, and leverage cloud models for highly complex tasks and extensive world knowledge.
ambient_temp_xeno@reddit
I think maybe that ferret licence is for the python code or whatever they've done. They're probably contractually obliged because as employees Apple probably owns anything they make on company time.
Ok-Creme-5377@reddit
not just on company time, as long as they're working there
lmamakos@reddit
Seems like you've never worked for some very large company with the associated painful processes. You're basing this strategic direction theory of yours by which GitHub repo some Apple internal group (among dozes/hundreds) published their software into?
nderstand2grow@reddit (OP)
Enlighten us then...
cyan2k@reddit
Well, 5 months later and it's still going strong.
16k stars, 1k forks, an absolutely amazing community, and good management in terms of feature planning and organization. It's one of the better managed open-source projects I have been active in over the past 20 years.
It seems very healthy and is only going to get better even though it has come a long way already. Performance-wise, it's almost on par with llama.cpp, and with swapping, you can even use 70B Q8s.
Also, it's really nice to actually program with it. More coherent and straightforward, because instead of needing to know your way through numpy, pytorch, diffusers, transformers, and whatnot, you only need to know mlx. Very cool.
nderstand2grow@reddit (OP)
I agree. I was wrong before. Working with it now and I like it. It still has a long way to go as many things are missing, but it's gota good head start.
a_beautiful_rhind@reddit
Interesting theory. The code is out there already and anyone can take over. I don't think apple is going to be chasing future MLX devs for maintaining it.
That said, losing official updates would suck. They're also free to deprecate in the same way that nvidia or AMD does. They just drop support for your chip when they come with something new and bid you good luck.
So far there are only 4 devs : https://github.com/orgs/ml-explore/people
nderstand2grow@reddit (OP)
Yeah, people here are downvoting me but they'll understand in a few months...
The 4 people you mentioned are all "research" scientists at Apple, they're not part of the products line. MLX for them is a curiosity, nothing more.
scapocchione@reddit
Few months have passed, man. The framework is becoming more and more complete, and the MLX team is hiring (Awni Hannun post on X).
nderstand2grow@reddit (OP)
is Apple actually using it in any product?
scapocchione@reddit
I don't really understand what you mean. Is Pytorch "used in any product"?
nderstand2grow@reddit (OP)
yes, Meta uses it extensively
scapocchione@reddit
You didn't catch the meaning of my reply.
Let's try again. Which Meta product(s) is Pytorch used in?
a_beautiful_rhind@reddit
The way they had it set up originally, before mlx, was to give you their models to use in your apps; that you got a developer account for, etc.
So sure.. apple is all in on ML but that ML is not running llama, it's integrating ML into their ecosystem.
People paid $6k for a computer though.. they are of course not going to be happy.
nderstand2grow@reddit (OP)
This is what people here struggle to understand (didn't know we had this many Apple fanboys here...)
I'm typing this on a MacBook Pro with M1 Pro chip, and I get access to M2 Ultra. But I also see the serious drawbacks of Apple's ecosystem and business model when it comes to LLMs.
JacketHistorical2321@reddit
I get roughly 20-24 t/s with a M1 ultra /128gb ram from a 70B 4q so no idea what you are talking about here
a_beautiful_rhind@reddit
Through mlx? L-cpp speeds weren't that high last I checked, not on a 70b. Because you're saying they're higher than GPU and that sounds like a lie.
scapocchione@reddit
Have you noticed 4q?
scapocchione@reddit
That's the entire point.