NEW: Meta Mark Zuckerberg Llama 3.1 405B, 70B, & 8B Model Release SHOCKS Industry! 🤖 OPEN SOURCE LLM

The first time we're releasing a 405 billion parameter model by far the most sophisticated open source model that that I think anyone has has put out Mark Zuckerberg at meta just released llama 3.1 and it is taking the industry by storm let's dive right into [Music] it all right so for those of you who don't know meta just released llama 3 . 3.1 and it is their most capable model to date and it's on par with some of the top closed

Source models like open AI GPT 40 as well as Claude Sonet 3.5 and we can see here some key takeaways um meta is committed to openly openly accessible AI uh you can read Mark Zucker Bird's letter detailing why open source is good for developers good for meta and good for the world I watched that interview of Zuck kind of I mean I've watched many interviews but his most recent one about the meta uh llama release he says a lot of stuff

That he said in many other videos regarding open source and whatnot but definitely a great video I would definitely recommend checking that full length 31 minute 35 minute video out um bringing open source intelligence to all our open our our latest models expand context length to uh 128k add support across eight languages and include llama 3145 billion the first Frontier level open source AI model so this is huge all right because up until this point open source has always been lagging

Behind and you know not a huge amount but still a substantial amount and now where where we have it where this is the first Frontier Model um that is getting close or on par with these other top Frontier models and this is 100% open source you can literally get the code on you know whether it's hugging face GitHub Etc um you can get the code you could if you have a strong enough system you could host uh it locally um depending

On you know what size this is obviously a huge model 45 billion parameters but they have smaller ones as well and this is yeah this is really game changing in the AI industry because as we're going to start seeing in the next few months and what we're already seeing now is these open source models are going to basically I mean what we're seeing right now they're they're on par if not better than a lot of the other Frontier models so it's

Only going to get go up and up from here so imagine in 6 to 12 months where open source is going to be um we're continuing to build on out uh LOM uh El Lama to be a system by providing more components that work with the model ining a uh reference system we want to empower developers with the tools to create their own custom agents and new types of agentic behavior we're bolstering this with new security and safety tools including llama

Guard 3 and prompt uh prompt guard to help build responsibly we're also releasing a request to comment on llama slack stack API uh uh a standard inter face we hope to make it easier for third party projects to leverage llama Models All right so obviously you know with Zuck uh releasing this they're going you know yes he's essentially pcking these millions billions of dollars they're putting into Ai and and and building these models and you're essentially giving it to the public

Making the source code available but at the same time obviously there's there's many different benefits for for meta um but one obviously being people are going to build on top of the open source project and that's Obviously good for them um you know if if their models are being used for other people's stuff the ecosystem is primed and ready to go with over 25 Partners including AWS Nvidia data Breck Gro Dell Azure uh Google cloud and snowflakes offering Services one uh

Day on day one all right so uh we'll get into a little bit of using the model in just a second and I'll do more videos you know going more in depth on testing llama 3.1 but um it's going to be really interesting once Gro like it's available with Gro because if you don't know about grock it's very essentially it's just uh very very fast um inference time so uh yeah I would definitely recommend checking out grock regardless um because it's

Definitely a very useful tool and um you know they use different types of uh lpus instead of gpus uh try llama 3.1 45b in the U us on WhatsApp and atm. by asking a challenging math or coding question so today open source large language models have mostly trailed behind their closed counterparts when it comes to capability and performance now we're ushering in a new era of Open Source leading the way we're publicly releasing meta llama 3.1 405b which we believe to

Be the world's most capable and openly available foundational model with over uh with more than 300 million total download of all llama versions today we're just getting started all right so I'm not going to go through and read this whole thing I'll have this Linked In the description but as you can see here um as part of this release date we're introducing upgraded versions so they're also upgrading the versions of 8B and 70b models these are multilingual and have significantly longer

Context lengths of 128k state-of-the-art tool use and overall stronger reasoning capabilities this enables are latest models to support Advanced use cases such as long-term text summarization multilingual conversational agents and coding assistance we're ALS we also made changes to our license allowing developers to use the outputs from llama models including the 405b to improve other models true to our uh commitment to open source starting today we're introducing these models available to the be downloaded true to our commitment to open source starting

Today were f making these models available to the community for download on l.a. comom at huggingface and available for immediate development on our broad ecosystem of partner platforms so some of the model evals here we can take a look for release we evaluated performance on the uh over on over 150 Benchmark data sets that span a wide range of languages in addition We performed extensive human evals that compare llama 3.1 with Compu models in real world activity or scenarios our experimental

Uh evaluations suggests that our Flagship models is competitive with leading Foundation models across a range of tasks including gp4 GPT 4 GPT 40 and Claud sunet 3.5 additionally our smaller models are competitive with closed and open models that have a similar number of parameters so we can take a look and see here um these benchmarks so 3.1 405b um and we can see some of the comparisons here against Omni and Sonic so on the general we're I mean like not to

Go through all these evals but if you really just look here um it's pretty hand uh head-to-head with Omni and Sonet on like e i if eval it is ahead and then five shot coot it is a little bit it's basically on par with gbt 40 Omni and just below son it um but yeah you can see here the ones where it's leading it will be uh kind of shaded so you can see math it really shines right here 96.8 and

CLA on it 96.4 uh Arc challenge so 96 .9 um Nexus so you can see here it's 1 2 3 4 5 6 7 1 two 3 four five 6 78 so there's uh 15 right here and out of 15 according to their Tas it got seven out of them um basically top performing out of all these models for this one and then you can see more tests here uh what is this this is their 8B model um so now they're

Comparing it to uh Gemma 2 mistel 7B instruct llama 3.1 70b and okay and then so there's their 8B model here and their 70b model here against going against mistol and GPT 3.5 turbo um so you can see the evals according to them are putting their new llama models ahead in a lot of these evals so um and then llama 3.1 405 human eval uh win tie loss so you can see the win and tie and loss rates and then model

Architecture um yeah so I'm not going to go through the entire um article here but I'll leave it linked below um the main thing I wanted to cover was kind of just you know um The evals and the this this huge release this is going to be gamechanging for uh the AI industry um and just really with open source On The Rise um you know with these open source models it's like I if you watch the interview with Zuck he basically

Is just saying a lot of companies are a lot of people are want going to want to fine-tune their own models and build their own models and just just having being at the whim of like a bigger company maybe like Claud Sonet or like anthropic or open AI um and not being able to kind of build what your own uh type of uh model and agent um is obviously what people are going to want so and yeah this is going to

Be uh interesting to see how open source continues to evolve let me know what your thoughts are in the comments down below regarding all this regarding llama uh 3.1 um if you use and also too guys you can go to their I'll leave a link to the meta AI right here and you can basically start talking to um you know their llama 3.1 model okay please create a blog post regarding open source AI advance so I'll be doing more videos and

Tests on this new uh meta AI but um yeah you can go test it out yourself as well and um yeah we definitely be interesting to see and start incorporating it with some um different tools trying it with coding and um more stuff to come so other than that guys I hope you got some value from this video you're new to the channel make sure to smash that like button smash that subscribe button also too if you haven't already joined our

Free stride Community stri community.com Link for that will be in the description down below we have likeminded entrepreneurs AI enthusiasts developers coders um business owners marketers salespeople Etc all in there so you can network with like-minded individuals as well as myself and then also too guys if you haven't already checked out my book client acquisition code link for that will be in the description Down Below tooo guys other than that I will see you in tomorrow's video keep hustling keep grinding

And of course accelerate your stride take care