Maximilien Moulin

How fast is fast when it comes to commercial open source software traction? Here are some stats to serve as benchmarks that Matthieu Lavergne put together among many other great data and insights in Serena's first report on commercial open source software. It's available for free here: https://lnkd.in/d5uBvjAU

18

3 commentaires

Meta has released Llama 3 (backed by a $10 billion investment) and is available to public for FREE 😲 . What could be the business strategy behind it 🤔 ? It could probably: 1.     Position Meta as a leader in open source. 2.     Create potential hurdles for smaller AI developers. 3.     Make it harder for companies like OpenAI to charge a premium. 4.     Drive adoption of its platforms by integrating with Instagram, Whatsapp, and Facebook (be ready for lots of Adsssss and suggestionssss….) It’s worth remembering the old wisdom: If the product is free, you might be the product! 🙄 #Meta #Llama3 #AI #OpenSource #BusinessStrategy #TechnologyNews

17

2 commentaires

Can you be someone as honest as the harsh reality, politically correct, fun at the same time, and also one of the most knowledgeable in artificial intelligence? Only if you are French and your name is Yann LeCun ! Take a look to this master piece from Yann 👇 <sarcasm> Come work at ClosedAI. With AGI just around the corner, your shares will be worth 42 sextillionnollars. We can claw back your vested shares if you quit, unless you sign a non-disparagement agreement. Oh wait, sorry, we didn't realize our contract was this harsh until someone published an article on it. You can keep them. We're worth 42 octillionnollars, but since we're a private company, you can't really sell your shares. Also, we can actually stop you from selling them. And you can't say a word about what it is we're working on. If you do, we'll fire you within the hour, claw back your shares, and you won't be able to say *anything*. Don't take it personally but, you know, we're soooo far ahead of everyone else and AI is soooo dangerous in the hands of the unwashed masses. It's safe only if *we* do it. </sarcasm>

35

10 commentaires

Wth are LLMs - Claude, LLaMA and of course, the more popular GPT models ? What happens inside these black-boxes ? How are they similar, and how do they differ from each other? So we look into: GPT – Leveraging transformer architecture, GPT-3.5 and GPT-4 power ChatGPT, honed through a mix of supervised and reinforcement learning for contextually rich text generation. LLaMA – Meta AI's LLaMA models, including LLaMA 2, offer a fresh take with leaner design, open-source accessibility, and potential for specialized language tasks, all runnable offline. Claude – Pioneering "constitutional AI," Claude ensures output relevance and safety by adhering to predefined principles, setting a new standard for practical and ethical text generation. We also look into the various tools being used for content generation, you can explore many of these tools and might find them very useful: - Text generators, conversational bots - AI-enhanced artistic creation - Dynamic audio synthesis, - A few multimodal synthesis tools The generative playground is expanding, promising a revolution in content creation across diverse mediums. ______________________________ To read more, please find substack link in comments A LOT of work goes into researching, studying, and articulating all this, hoping you will find it useful. #GenerativeModels #AI #LLMs

34

2 commentaires

Already 8 months when i presented autogen V.02for Gen Ai startup program with Microsoft incubator startup n STATION F and now we are here. ❤️ This library has demonstrate how to use #agents / multi LLM - SLM with decision pattern ( tooling RAG, script , code ,apis ) to adress usecases and pipelines of creation of business values. ( and not only for a chatbot...) Autogen rises ( a lot ! ), new capabilities, low-code studio appears... Framework orchestration in python, java, c#... demonstrated how to create application with #AI. 🔥 Gen AI is obviously mature for business and pratical usecases are already in production.🚀 🏃 And if you don't know about #Autogen ? >> Rush into youtube / website to understand why it must be a must known for your business 🔥 One of the next step ( low code for autogen Studio ): https://lnkd.in/ea64EF5b the library : https://lnkd.in/e4W7_gHT 📜 First demo live ( french) : https://lnkd.in/eTchhs5n 📜 First demo presential :https://lnkd.in/ehmjwAgZ #agents #microsoftfrance #llm #slm #genai #ai #autogen

28

// AI Generated Code that's Nearly Impossible to Debug // "Now, thanks to ChatGPT, you could improve this workflow by: - making step 3, the only easy step, somewhat easier (and thus hiring less competent engineers) - making step 5 nearly Impossible."

1

💥 New day, new model 💥 We ran our performance benchmarks on Anthropic Claude 3.5 Sonnet v/s OpenAI's GPT-4o & the results weren't what you'd expect ⏬ Anthropic says Claude 3.5 Sonnet outperforms all competitor models on key evaluations, at twice the speed of Claude 3 Opus and one-fifth the cost. In our benchmarks 3.5 Sonnet & 4o came out neck & neck. Ultimately, you need to run evaluations on your task at hand and measure the quality yourself. 🔍 Data Extraction: Both models identified 60-80% of data correctly, but neither excelled in this task. 📊 Classification: Claude 3.5 Sonnet (72%) outperformed GPT-4o (65%) in mean accuracy. However, GPT-4o led in precision (86.21%), crucial for accurately classifying customer tickets. 🧠 Verbal Reasoning: GPT-4o outperformed Claude 3.5 Sonnet with 69% accuracy, versus 44%. Both models did well on analogy and relationship questions. 📜 Context window: Claude 3.5 Sonnet boasts a larger context window (200K) compared to GPT-4o's 128K 🌪 Latency: Claude 3.5 Sonnet is 2x faster than Claude 3 Opus, but it’s still lags behind GPT-4o when it comes to latency (50% faster Time to First Token) Price and throughput for these models was very similar Want to learn more about how these models stack up? Check out our full analysis on the Vellum blog: https://lnkd.in/dwEcQJe4 Oh and did I say, Claude 3.5 Sonnet is already in Vellum? Sign up for a demo to run your evals & start building!

67

1 commentaire

Totally agree, you can see the censorship of what gets published to whom in all of the social media platforms. Take it to the next level and let the AI figure out an optimize the revenue and/or political agenda and you have total control over each person‘s view of reality. This is one of the reasons why I use multiple browsers and multiple social media accounts when I want to see different things based on what the AI systems know or don’t know about me. I think this is a more likely scenario than a full-blown existential threat like war between AI and humans. The ability to persuade people and get them to do what you want without them knowing that you were doing it is the likely scenario. Analogously just like the brain (AI) using the nervous system (the network) controlling an arm (the group of people to be influenced) to pick up an apple (the task that needs to be persuaded). Let’s face it. Honestly we are literally just cogs in the wheel or cells of an appendage.

2

Forget about LLMs Ranking Leaderboards🥇 , Welcome to the Era of Model Routing 🔀 ! The more I talk to AI practitioners, the more I realize about a clear shift in the market. Few months ago there was no discussion about which model to use. Today, things are very different. Talking to a customer last week, he said: I like Claude 3 for long summaries, GPT4 for coding tasks, Mistral 7B for simple user queries and the list continued….. Imagine choosing a foundation model like picking a car 🚗. You want something cost-effective, reliable, and versatile. But sometimes, you need the speed of a Ferrari 🏎. Easy choice, right? Not so fast. Traditional model selection has been static, tied to specific use cases. What if your journey includes a mountain track? Suddenly, you wish you had a Range Rover instead. Picture this in the world of generative AI. You've got a complex technical report for analysis and then need to send an email summarizing it moments later. You don't necessarily need the same model for both tasks—just the right one for each specific job. That's the beauty of model routing. It dynamically selects the most suitable model based on the task at hand. As companies expand their use cases, I foresee model routing becoming increasingly crucial, especially with the advent of state-of-the-art models and others fine-tuned for specific tasks. The future of model selection is flexible and task-specific, marking a significant shift from how we've approached it until now. 𝘋𝘰 𝘺𝘰𝘶 𝘸𝘢𝘯𝘵 𝘢 𝘳𝘦𝘢𝘭 𝘦𝘹𝘢𝘮𝘱𝘭𝘦 𝘰𝘧 𝘮𝘰𝘥𝘦𝘭 𝘳𝘰𝘶𝘵𝘪𝘯𝘨? 𝘛𝘩𝘪𝘴 𝘪𝘴 𝘩𝘰𝘸 Amazon 𝘘 𝘧𝘰𝘳 𝘉𝘶𝘴𝘪𝘯𝘦𝘴𝘴 𝘮𝘢𝘯𝘢𝘨𝘦𝘴 𝘢𝘯𝘺 𝘶𝘴𝘦𝘳 𝘲𝘶𝘦𝘳𝘺. 𝘐𝘵 𝘶𝘯𝘥𝘦𝘳𝘴𝘵𝘢𝘯𝘥𝘴 𝘪𝘵𝘴 𝘤𝘰𝘮𝘱𝘭𝘦𝘹𝘪𝘵𝘺 𝘢𝘯𝘥 𝘤𝘩𝘰𝘴𝘦 𝘵𝘩𝘦 𝘮𝘰𝘴𝘵 𝘴𝘶𝘪𝘵𝘢𝘣𝘭𝘦 𝘮𝘰𝘥𝘦𝘭 𝘪𝘯 𝘵𝘦𝘳𝘮𝘴 𝘰𝘧 𝘱𝘦𝘳𝘧𝘰𝘳𝘮𝘢𝘯𝘤𝘦 𝘢𝘯𝘥 𝘴𝘱𝘦𝘦𝘥, 𝘵𝘳𝘺𝘪𝘯𝘨 𝘵𝘰 𝘮𝘢𝘹𝘪𝘮𝘪𝘻𝘦 𝘵𝘩𝘦 𝘶𝘴𝘦𝘳 𝘦𝘹𝘱𝘦𝘳𝘪𝘦𝘯𝘤𝘦. #ai #genai #llm

321

34 commentaires

LINAGORA: France 2030 Winner as part of the OpenLLM-France project: A victory for Open Source Generative AI A look back at a key day bringing together the AI Champions 🇲🇫 at the Elysée Palace on May 21st... Told by our President on his personal blog ✨ Our OpenLLM-France project is proof that an alternative, truly free, and sovereign model is possible. We are committed to creating generative AI that is trusted, open, and accessible to all. ✨ Alexandre Zapolsky 1️⃣ A Vision for the Future The National Strategy for AI (SNIA), initiated in 2018 and strengthened in 2022, aims to spread AI in the French economy to improve industrial competitiveness and collective well-being. Three main levers are activated: ➡ Training and research: Strengthening institutions of excellence with €560 million in public funding; ➡ Support for a cutting-edge offer: Development of cutting-edge AI technologies; ➡ Bridging supply and demand: Facilitating the adoption of AI by businesses and public sectors; 2️⃣ A visionary project Our OpenLLM-France project aims to develop fully Open Source generative AI models, trained on high-quality academic data in French. We aim to create an API and chatbot for education, leveraging Open Educational Resources (OER). This initiative will support National Education and the EdTech industry to offer personalized and interactive learning paths. 3️⃣ An Alternative to the Tech Giants Faced with the concentration of technology giants and their overcapitalization strategies, OpenLLM-France is positioning itself as a collaborative, open and industrial alternative. The full text can be found here on his blog: https://lnkd.in/e9J59H88 w/ OpenLLM 🇫🇷 🇪🇺 opsci TALKR Commissariat a l'Energie Atomique et aux Energies Alternatives CNRS Loria University of Paris I: Panthéon-Sorbonne #DigitalSovereignty #AI #LLM #opensource

5

Better & Faster Large Language Models via Multi-token Prediction Fabian Gloeckle, Badr Youbi Idrissi, Baptiste Rozière, David Lopez-Paz, Gabriel Synnaeve Abstract Large language models such as GPT and Llama are trained with a next-token prediction loss. In this work, we suggest that training language models to predict multiple future tokens at once results in higher sample efficiency. More specifically, at each position in the training corpus, we ask the model to predict the following n tokens using n independent output heads, operating on top of a shared model trunk. Considering multi-token prediction as an auxiliary training task, we measure improved downstream capabilities with no overhead in training time for both code and natural language models. The method is increasingly useful for larger model sizes, and keeps its appeal when training for multiple epochs. Gains are especially pronounced on generative benchmarks like coding, where our models consistently outperform strong baselines by several percentage points. Our 13B parameter models solves 12 % more problems on HumanEval and 17 % more on MBPP than comparable next-token models. Experiments on small algorithmic tasks demonstrate that multi-token prediction is favorable for the development of induction heads and algorithmic reasoning capabilities. As an additional benefit, models trained with 4-token prediction are up to 3 times faster at inference, even with large batch sizes. 👉 https://lnkd.in/dXSXRErV #machinelearning

1

𝗖𝗧𝗢𝘀: how can you tell if the use case that Product dreams about is 𝗮 𝗴𝗼𝗼𝗱 𝗳𝗶𝘁 𝗳𝗼𝗿 𝗔𝗜? 🤔 The potential of Transformers, the architecture at the core of the famous LLMs, is huge and so is the potential of over-engineering. ⚠ “Can AI do this?”, “can AI do that?” is how most business approach the issue. The answer is almost always “yes” but the real question is “𝗵𝗼𝘄 𝘄𝗲𝗹𝗹?” To understand what LLMs can do, 𝘄𝗲 𝗳𝗶𝗿𝘀𝘁 𝗻𝗲𝗲𝗱 𝘁𝗼 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗰𝗮𝗻𝗻𝗼𝘁 𝗱𝗼. So let’s break it down to first principles 🤓: We can break down Software Engineering to 𝘀𝗲𝗾𝘂𝗲𝗻𝗰𝗲𝘀, 𝗰𝗼𝗻𝗱𝗶𝘁𝗶𝗼𝗻𝘀 𝗮𝗻𝗱 𝗹𝗼𝗼𝗽𝘀. That's the minimum necessary to instruct a machine. Similarly, we can break down LLMs to their atomic skills too. LLMs are great at learning 𝗰𝗼𝗺𝗽𝗹𝗲𝘅 𝗽𝗮𝘁𝘁𝗲𝗿𝗻𝘀 and applying those patterns to data that they have not "seen" before. 𝗕𝘂𝘁 𝘁𝗵𝗲𝘆 𝗱𝗼 𝗻𝗼𝘁 𝘁𝗵𝗶𝗻𝗸, 𝘁𝗵𝗲𝘆 𝗱𝗼 𝗻𝗼𝘁 𝗽𝗹𝗮𝗻 𝗮𝗻𝗱 𝘁𝗵𝗲𝘆 𝗱𝗼𝗻’𝘁 𝗵𝗮𝘃𝗲 𝗮𝗻 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗼𝗳 𝘁𝗵𝗲 𝗻𝗮𝘁𝘂𝗿𝗮𝗹 𝘄𝗼𝗿𝗹𝗱. ❌ They simply generate the next token from a 𝗽𝗿𝗼𝗯𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 to follow a learned pattern: summarise, extract, translate, analyse, write, transform to JSON, etc. 📊 ChatGPT doesn’t plan a city break for you because it’s intelligent, but because it was trained on patterns of human communication and it can associate Rome with the Colosseum because it has seen the two together in a lot of texts. 🐒 More on how they actually work in a future post. 😉 The rules of the game have not changed: 𝗶𝗻𝗽𝘂𝘁 𝗰𝗿𝗲𝗮𝘁𝗲𝘀 𝗼𝘂𝘁𝗽𝘂𝘁, or completions in this case, except you don’t define every execution step in code, but you exploit a learned pattern. 𝗦𝗼, 𝘁𝗼 𝗮𝗻𝘀𝘄𝗲𝗿 𝗼𝘂𝗿 𝗶𝗻𝗶𝘁𝗶𝗮𝗹 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻, 𝘄𝗲 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗰𝗼𝗻𝘀𝗶𝗱𝗲𝗿 𝘁𝘄𝗼 𝗮𝘀𝗽𝗲𝗰𝘁𝘀: 1️⃣ Can this problem be easily solved with sequences, conditions and loops? If yes, AI might be overkill. 2️⃣ Do I have an LLM that knows this pattern and can execute this step of the flow consistently and predictably? If not, can I build one? (check out my other post on how much that costs) The LLM is just another service in your tech stack that enables a user journey. 𝟴𝟬% 𝗶𝘀 𝘀𝘁𝗶𝗹𝗹 𝗰𝗹𝗮𝘀𝘀𝗶𝗰 𝘀𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗮𝗻𝗱 𝗵𝗮𝗻𝗱𝗶𝗻𝗴 𝗼𝘃𝗲𝗿 𝘁𝗼 𝘁𝗵𝗲 𝗻𝗲𝘅𝘁 𝘀𝗲𝗿𝘃𝗶𝗰𝗲. LLMs are a great technology that allows us to tackle problems where sequences, loops and conditions would have been too complex to write and maintain and where following patterns learned from data can deliver consistent, repeatable results. The world is changing and staying the same at the same time. ⏱ Happy building! 🔨 👋 ------------------------ I share AI content from my 𝗱𝗶𝗿𝗲𝗰𝘁, 𝗵𝗮𝗻𝗱𝘀-𝗼𝗻 𝗲𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲. Please 𝗰𝗼𝗺𝗺𝗲𝗻𝘁 💬 , 𝗹𝗶𝗸𝗲 👍 𝗼𝗿 𝗿𝗲𝘀𝗵𝗮𝗿𝗲 ♻ if you think others would find it useful. 🙏 Hit the 🔔 to stay updated.

38

9 commentaires

🚀 𝗘𝘅𝗰𝗶𝘁𝗶𝗻𝗴 𝗡𝗲𝘄𝘀 𝗳𝗿𝗼𝗺 𝗠𝗲𝘁𝗮! 🌐 Meta is pushing the boundaries of AI innovation with the launch of Llama 3, our latest version of the foundational Open Source model. 🦙✨ 𝘓𝘭𝘢𝘮𝘢 3 𝘪𝘴 𝘯𝘰𝘸 𝘢𝘷𝘢𝘪𝘭𝘢𝘣𝘭𝘦 𝘪𝘯 𝘵𝘸𝘰 𝘳𝘰𝘣𝘶𝘴𝘵 𝘷𝘦𝘳𝘴𝘪𝘰𝘯𝘴: 8𝘉 𝘢𝘯𝘥 70𝘉, promising performance that rivals or surpasses similar models in the market. 🏆 🌟 𝗕𝗶𝗴 𝗥𝗲𝘃𝗲𝗮𝗹: Get ready for our most ambitious version yet - a 400B multimodal model, set to revolutionize how we interact with AI by supporting both text and image inputs! 🎨📝 Meta AI has officially gone web! You can now interact with our AI chatbot directly from your browser at www.meta.ai - No sign-up required! 🖥️💬 This move marks a significant milestone as we continue to enhance user accessibility and push the envelope in AI technology. 🔍 𝗗𝗶𝘃𝗲 𝗶𝗻𝘁𝗼 𝘁𝗵𝗲 𝗳𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝗔𝗜 𝘄𝗶𝘁𝗵 𝗠𝗲𝘁𝗮 𝗔𝗜 𝗮𝗻𝗱 𝗟𝗹𝗮𝗺𝗮 𝟯 - whether it's coding assistance or multimodal interactions, they're here to provide an unparalleled experience. #MetaAI #Llama3 #AI #Innovation #TechNews #ArtificialIntelligence #OpenSource #Lognext Lognext

10

I co-hosted an AI engineering leaders dinner on Thursday with Grace Ge. Thank you to Beyang Liu from Sourcegraph, Devendra Chaplot from Mistral AI, Bryan Bischof from Hex, Marco Sanvido from Robust Intelligence, and Tim Shi from Cresta for joining us. Here are the a few key takeaways: 1️⃣ AI hardware has a long way to go, but folks in the community are generally excited about how it can influence our day-to-day. Right now, they're toys showing us glimpses into the next generation of consumer tech. I respect rabbit inc. / Humane for starting to make progress in this area, but I think there's a lot more to come 😁 2️⃣ We're at an interesting crossroads between engineering and research in the AI community. There's a lot of respect on *both* sides, but I find that these conversations often pendulum swing between specific research areas and very pragmatic scaling/tuning insights. This paper on Chinchilla by Rohan P. came up specifically as important new work: https://lnkd.in/gD6tmDRv 3️⃣ Everyone feels that AI can help them get out of the busy work, and help focus them on the high-leverage parts of their jobs. There does seem to be a tension, though: those solutions are still "not quite good enough" but the ones that make for cool demos look like problem-seeking solutions. The products mentioned were Granola and Agenda Hero. 4️⃣ Large categories of companies within AI are already being commoditized quickly. There was a lot of talk about which companies are taking the "smart bets" in some of these competitive categories, but it feels to me like these 5-7 year strategies will start paying dividends and creating category kings within the next year or so. Specific excitement around players like turbopuffer. 5️⃣ Data privacy remains a hot topic. It's not clear if the incentives line up to push in the privacy direction for most large companies, and I think a lot of R&D cost is being driven by what AI talent *wants to work on* rather than productionizing the whole ecosystem. Thanks again to Grace Ge for the wonderful venue and Sunil Dhaliwal + the Amplify Partners team for co-hosting with Orb!

72

7 commentaires

4o ? More like 4oh-no amirite? The latest model from OpenAI, named 4o (o is for "omni") can do so much that that it might make you think "oh no, I'm about to be replaced by an AI." The word "awesome" really fits for describing the new capabilities and what this advancement could mean for human life as we know it. It's one thing to get text or a picture generated, but the 4o model can joke with you, help you get ready for an interview, or tutor your children! It can also read facial expressions, make inferences based on context, and offer words of encouragement. So since this new virtual personal assistant can do everything... what are we supposed to do now? Are teachers replaced? Writers? Stylists? No job is safe, because the computer can do it all. Not quite. The computer cannot care. It's wild that I have to say this - I thought it was only stuff for sci-fi, but here were are. Computers do not care. They cannot express compassion. They cannot empathize. They can simulate these things but cannot perform them, because they are not human. All of the advancements we make in technology are nothing but cold and heartless science if we don't take into account that the needs being addressed are human needs. Humans can care. We're wired for it from the ground up - if we see someone in pain, sometimes we can feel it even though we're just seeing or hearing about it. We are fallible, imperfect, even a bit messy - but we can offer true compassion. So... do you really want to get coaching and tech help from a machine? How many youtube videos can you watch about a topic before you start to wonder "if I could just ask this person a simple question I'd understand this so much more"? Even a talking computer assistant can only regurgitate what it has been fed, even if may be able to rephrase and summarize complex content all on its own. It cannot understand your motivation, it can't empathize with your lived experience as a person. These things matter! If you'd like help getting over the limiting beliefs that are holding you back from being a confident and competent user of your day-to-day business tech, and you're ready to work with a human who cares, book a call with me today! https://askmaria.tech

4

Deciphering AI's Inner Workings: Scaling Sparse Autoencoders to Claude 3 Sonnet - A good read! https://lnkd.in/eYuwg7_U

9

BIG NEWS ! Today, Google Cloud and Anthropic made some exciting announcements: 📣 Claude 3 Opus, Anthropic’s most intelligent model, is now generally available on Vertex AI ⚙️ Tool use is available for the Claude 3 model family on Vertex AI, enabling Claude to act as an autonomous agent 📈Provisioned throughput is available for the Claude 3 model family on Vertex AI

17

1 commentaire

Presenting Orpomis Prime 7B model series ! These models are a result of trying different merge strategies of the following models: - mistral-orpo-beta from KAIST AI - Hermes-2-Pro-Mistral-7B from NousResearch In my experiments, dare ties merge leads to a significantly better-performing model on the openllm benchmarks In general, the model performance is quite solid in text generation as well as math reasoning. Full result comparison: https://lnkd.in/gb92DXwG Dare ties Model: https://lnkd.in/gMS-H4Sq Model stock Model: https://lnkd.in/giwEJ7Gs

9

🦙 Meta AI Releases Llama 3: 8B & 70B Models! 🚀 Meta has released two new Llama 3 models : 8B parameter model outperforms 70B Llama 2 🤯 70B parameter model is competitive with Gemini Pro 1.5 and Claude 3 Sonnet 💪 (more models are coming soon!) 405B model in the works, nearing GPT-4 level 👀 🔑 Key Takeaways: Trained on 15T tokens, 7x more than Llama 2 8K context window, shorter than some recent models Performs well on reasoning benchmarks like GSM8K Supports commercial use with some license restrictions 🧪 Trying it Out: Available via Ollama, HuggingFace, API services Can be run locally with quantization Responds well to chain-of-thought & ReACT prompts Some multilingual support but dedicated model likely coming 📜 License Notes: - Can't use Llama 3 to train other large models - Must include "Llama 3" in name of fine-tuned versions - Review prohibited use cases before commercial use The Llama 3 models look promising and signal continued rapid progress in open LLMs. Expect to see powerful fine-tuned versions and new capabilities soon! 🦙🔥 #Llama3 #OpenLLMs #LLMProgress

74

2 commentaires