Hello community, today we have two brand new AI models, and you're not going to believe it - they are French and European! Here we have it: a new Pix2Large, a 124 billion free, trainable parameter, open-weight multimodal AI, a vision-language model right out of Paris. Isn't it beautiful?

And yes, there's one particular test. This is called 'Mathematical Vista,' which evaluates complex mathematical reasoning over visual data. This model outperforms all other models on this planet. Isn't it beautiful? So, we also have, as I told you, an update here

to the Mr. Large. It has a maximum token of 128k, and the version is 24.11 for 2024 November. Plus, our new multimodal model, and this is called Pix2Large, also version 24.11. Now isn't it beautiful? You have

your API for texture generation, vision code generation, embedding, function calling, fine-tuning, JSON mode, and God railing. Plus, you have beautiful developer examples. You want to build your RAG for internet retrieval, RAG for the medical domain, a financial advisor, or a legal co-pilot? You have all those

models available here on the documentation and on the platform. You can implement it, and on Hugging Face, we can explore it, and we will do this. But at first, you want to build an agent here with this latest model. And I just checked it in - yes, it is available. Have a look

here. You just go here on the platform to agents, you can create, you can deploy your agent with just a few clicks. You have here the model, and now finally, we have here the Pix2Large. This brand new model is only at 123 billion free

trainable parameter M, so we lost 1 billion, but never mind, it is here. European, you have the randomness, you have the instructions, you have the system prompt, you can have the ICL in context learning example, and you create an agent just by clicking here on deploy and this

agent is available on the complete platform, whatever you want to do with it, and you also have use cases. You have a French agent, and then you have a little bit of a Python agent, and you have a Python agent workflow and a data analytical multi-agent workflow and a

role-playing conversational agent. You just click on an AI, and it is available for you. Yes, today we are here promoting the European Union models today, but you're going to say, 'Hey, listen, I'm just interested in two points. Is it better than this Llama 3.29 billion?' Yes, look at

this. Here's our little Llama, and here is here the European Champion here from Paris here, and you also say, 'Hey, but is it as good as Claude 3.5 Sonnet version 2?' And well, look here, we found one test where it is. Yeah, it is better. I mean, you

could argue here with Claude 3.5 Sonnet here in this color that here Sonnet is a little bit better, but here the French model is better. Sonnet is better, the French model is a little bit better, so let's, it is minimum as good as Claude 3.5 Sonnet

version 2, and this is an emotional statement backed up by the data. Isn't this beautiful? Yeah, Advent is here. Christmas is coming, and we have a 128k context window and it fits a minimum of 30 finally high-resolution images, and yes, of course, it can generate

images and all the images you see here are generated by this new French model so if you want more here is the official link by Mool Pixol Large Pixol finally grows up isn't it beautiful and if you want to have the data here of all the

benchmark data here you see here in the first line Pix to Large 125B compared to the Llama 3.2 90B to Gemini 1.5 Pro to gbd4, Ser Omni, Claw 3.5, Sonic. This is version two, please note this is

version two, so also real good. Plus we have our text model, please do not forget we have here a beautiful large language model, only it's called Mr. Large 24.1. It's a significant upgrade to the latest version. We have notable improvements in long context

understanding. We have a new system ProMed and more accurate function calling. Isn't this beautiful? Plus this is highly capable for agent workflows for enterprises, your use cases as knowledge exploration, semantic understanding of documents, task automation. I say what semantic understanding of

documents? Let's test this out immediately since it's available now for 2 hours already here. So we go here to the chat platform and you see this is my prompt and I upload here the PDF file here and my prompter simply says, 'Hey, analyze the provided PDF document. You

have to objective and goals clearly define the primary aim of the research, discuss the specific problem in the eye that this research addresses, summarize the novel methodologies introduced, and highlight the most important facts in data. Explain how the methods differ from and improve upon existing approaches. Identify the main insights

and conclusions drawn by the authors discuss the relevance and the significance of these insights. Analyze how these new methods and findings integrate with current AI technology. Describe how this research advances the field of AI. Explore the potential implications and applications of this research.

Consider future research directions and practical applications that could arise from this finding and provide a concise scientific summary of the significance and the potential impact of this research on the broader AI landscape. Now to the reply, but this beautiful new model has 124 billion trainable parameters.

It comes back and says, "Here's a comprehensive analysis of this PDF about Graph Assistant, query and reasoning systems." Here's the primary aim, problem addressed, novel methods introduced, key methodologies, key findings. Second page here is then here improved accuracy, see explainable insights, significant main insights for the hybrid Rec system.

Page three goes on to integration in the AI landscape, broad implications, impact, potential implications. Page four comes to future research directions and then we have here the concise scientific summary in a very short form. So four pages here as a reply by this beautiful agent. I think it looks really, really good in the

performance plus. Then I can say, "Hey, can you find additional information on this topic on the internet?" And you say, "How is this possible?" Because look, it's nice graph, GraphQL, Microsoft Graph, new graph reasoning plus we have all the links to the original documentation on

the internet and it's easy because now we have beta features and I activated here the web search. But image generation? Yes, all images here in this video were generated here with this new M model, plus you have a canvas, you know the canvas already, that we can operate

when we do coding and whatever this is all available now and of course all of those models are available on Hugging Face and we go here to Mistral AI and as you see updated 2 minutes ago my goodness channel is perfect here we have our Pixol Large

Instruct 24B plus we have here the MRAI Large Instruct our text model 24B updated already an hour ago hey but here the New Vision Language Model brand new knew 2 minutes ago you just go there and you have all the information that you need

plus if you download here Zero hey it's 2 minutes already available what I wanted to show you if you go with the Vision Language Model with the Pixol 24B there is something beautiful because either you have here the installation pip install yes yes

server image beautiful and then you just ping the client and you have here the complete code you just copy the code request the Hugging Face up the daytime beautiful you go here with your model you have your load your system prompt great then you have the image URL then you

have the message and the text and everything and you get a response by the system this is great but you know what you have also you can ping here this Vision Language client also with a text only example and here you have the code in the following example shows how the

system prompt can be used to make sure for example that a model always knows the current date plus we're here in text only example so here you have the Python code if you want to implement this and you have here a simple role a simple question here as a user and then you see

Here's the response by the model and it knows the data and it knows exactly how many tests passed and I say this is a really nice feature. Offline examples, with everything is available to code for you, so I think there's nothing holding you back to enjoy it. Maybe you start here

with the research license and if you're in a company and you find it really helpful, well, contact the people at Mr. AI for a commercial license. So here we go. My last video, remember I showed you here uh, AI integration here, and I was talking here

that Meta published here that now the Llama model is available here for the US government and all services, civilian and non-civilian services, but today, today we have something beautiful because now if you look at this, America has with Meta, it's here, Llama 3.2 where

Meta declares this will be the new standard here, great, beautiful. We have in China here some beautiful, high-powered, impressive models like Q2.5, the complete family for text and vision and coding and everything, real good performance, and finally, starting tonight, starting tonight, we have in Europe also

a beautiful new LLM, a vision language model in addition, Mix 12 large version two. So, Asia now, as you can see, we are somehow starting to get balanced here in, at least if we look here at the LLMs and vision language models for the US, for

Europe, for China, this is it. This is just a first, very first impression here that I saw. Hey, we have a new European AI model and language model and vision model. I hope it was informative. I hope you enjoyed it. Try it out. The prices are available online. I was using here the research

license if you use it for commercial purposes, for commercial reasons you will get a commercial license. This is it for today. I hope you enjoyed it and maybe if you subscribe you will see me in my next video.

NEW VLM: PIXTRAL 124B 24.11 - Better Than Sonnet?

A pure research channel (R&D). Code new Artificial Intelligence (AI) models with @code4AI - where complex AI concepts are demystified with clarity grounded in theoretical physics. Delve into the latest advancements in AI, from transformative Visual Language Models and Graph-Based AI Systems to cutting-edge Text-to-AI Game and interactive Video Environments. 

Our explanatory content bridges the gap between theoretical physics, computer science, and AI, offering deep insights into Large Language Models (LLMs) to Vision Language Action (AI Robotics). Discover new mathematical models powering the core of AI. New AI World Models. 



NEW VLM: PIXTRAL 124B 24.11 - Better Than Sonnet?

NEW VLM: PIXTRAL 124B 24.11 - Better Than Sonnet?

Scripts