Apple beats EVERYONE to share LLM Secrets!!!
HTML-код
- Опубликовано: 5 сен 2024
- 🔗 Links 🔗
OpenELM: An Efficient Language Model Family with Open-source Training
and Inference Framework
arxiv.org/pdf/...
Apple OpenELM License - huggingface.co...
OpenELM Collection on Hugging Face - huggingface.co...
OpenELM FULL (CoreNet) Training and other Scripts - github.com/app...
❤️ If you want to support the channel ❤️
Support here:
Patreon - / 1littlecoder
Ko-Fi - ko-fi.com/1lit...
🧭 Follow me on 🧭
Twitter - / 1littlecoder
Linkedin - / amrrs
Just like Facebook, opensource is 1. a weapon against their competitors getting to far ahead, and also 2. A great way to create brand identity
Yes, all these companies are sociopathic. They never do anything for others unless there are anterior motives and the value to others is essentially nothing compared to what they think it gets them in return.
Apple open sourcing ANYTHING is a reason for celebration 🙌
It means its a worthless or already founded tech
@@riflebird4842 idk. You could be right. As far as already founded ...Seems apple did the hard work of doing r&d using public papers. The even use public datasets per this video. At the end of the day I think it's a good thing. And I commend Apple for spending the money and expertise doing the hard work. And making the knowledge publicly available.
Apple actually open source a lot of things :D
@@longmabaNo they don’t. Compared to how much is closed sourced their OSS offerings is almost nothing.
Apple deserves no praise. It’s a greedy, corrupt company that screws over consumers, lies with greenwashing tactics to pretend to care about the environment, disingenuously pushes far left political propaganda for PR reasons. Tim Cook is a scumbag and making some garbage LLL open source which so many others have already done, is just another meaningless PR stunt.
I request you buddy please make a beginner course on Langchain with ollama without any APIs, I don't know how to use Langchain. Please 🙏🙏❤
They are opensourcing what they need to run locally. After shipping a LLM to a device you basically leaked it. Open Sourcing offload part of the maitenance and make the company look cool.
Hey, in your next video can you cover, "Are Emergent Abilities of Large Language Models a Mirage?" Research paper by Stanford University presented at NeurIPS, it is very important for people to know that the emergent behaviours are just a result of evaluation and not something intrinsic to the model. Basically current LLM evaluations are not continous, answers are given a 0 or 1 for wrong and correct answer. But when the evalutaion is made to be continous in nature, the emrgent behaviour vanishes. I think the research paper is openly accesible for everyone to read and if you find it tough to understand just ask claude to explain and give examples. These emergent behaviours are just over hyped.
Thanks for the recommendation. Will check it out!
You argue that the paper’s premise is flawed because it doesn't fully grasp the concept of emergence. Your points include:
Emergence as a Definable Quantity:
Emergence involves a complexity that cannot be reduced to the properties of individual components.
It is characterized by a computationally irreducible barrier between two bodies, meaning the emergent behavior cannot be predicted simply by analyzing the components in isolation.
Paper’s Focus on Scaling:
The paper seems to focus on the scaling of models and the metrics used to measure performance, potentially missing the broader implications of what constitutes emergent behavior.
@@phobosmoon4643 did you just paste part of your conversation with an LLM?
@@veqv Yes. About a rhetorical argument. 2024
like always, you brought another informational video! pls show us, how to run the OpenELM-3B instruct model
Learning from meta, if you can't win it, open sourced it.
We don't see the datasets anymore because a "web" dataset is actually just unofficially-generated GPT-4 data -- and since GPT-4 is on the "web"... (you get the picture).
It will probably be Claude Sonnet 3.5 as the next "web" dataset though, haha. :)
Hello 1littlecoder,
I'm a big fan of your work
Please can you take a video on how to read and understand research papers as a beginner and also dsa for a junior ml engineer
❤❤
The paper says April, HF says "about a month ago", why this video now?
Thank you for this review! I am a bit disappointed that they did not benchmark HumanEval tho...
Hasn't Apple been working on this for a few months now? Did they just release it as open source? It's a significant step forward. It will take companies like Meta and Apple to keep opensource aligned and feature parity with Close source. Very important for AI to thrive.
You actually think any of these companies care about moving the tech forward unless it only benefits them? These open source models and these papers with them are just PR stunts and tactics to stifle smaller upstarts. They’re not much better than toys, they have very limited use.
¡Muchas Gracias!
Hey, doesn't that makes Pythia better as it was trained with less tokens(0.3T) compared to OpenELM (1.5T) [Although average accuracy is 4.1% more in the OpenELM]
You just have to look at the new macbook's specifications to realize why this makes so much sense for apple
what do you mean ?
please elaborate more
Gimme link to your discord server xD
don't have one (yet, maybe) ? :D
@@1littlecoder Interested in the discord server as well.
❤
I have request add Hindi audio track into your video, because my English is weak, with Hindi I can understand technical things easily
Maybe create a ai service that live translate audio for browser/phone audio. Samsung s21 phones does live transcriptions but I'm not sure about live translation.
Spoiler alert, do you know how to know when companies have the best LLM pipeline? When their model is better that the others..
whats the catch? There must be a catch with Apple? Surely thats not for charity
Sound like they're making it easier for app developer to build models. This will strengthen and differentiate apps on the apple store strengthening their foothold in smartphone market.
They want others to make apps that use these toy models so they don’t have to spend any money making their own. Why? Because then they can BS consumers into thinking they need their latest hardware to run these apps and make lots of money off a market others pay to build for them.
Then they can also screw devs over by charging their huge AppStore commissions when anyone does buy one of these apps while also using their restrictive AppStore policies to make sure nobody makes anything that actually competes with them.
Everything Apple does is self serving and shady.