Raahul Seshadri’s Post

I wanted a platform to support my #upskilling efforts in #GenerativeAI, #MachineLearning and #NaturalLanguageProcessing. Finding one that met my needs took me some time, labor and a surprisingly modest amount of cash. I built one myself with refurbished parts for under USD 500, and it performs fairly well with open source #LLM models. Click below to learn how I did this. #LargeLanguageModels #GenerativeArtificialIntelligence #ReduceReuseRecycle

Exciting news in the LLMs world: - 𝐌𝐢𝐬𝐭𝐫𝐚𝐥 𝐍𝐞𝐦𝐨 𝟏𝟐𝐁: Mistral released a new model in collaboration with Nvidia. 128k context window and multilingual capabilities (English, French, German, Spanish, Italian and more). Both base and instruct version available. Base MMLU 68.0%, instruct 53.4% MixEval Hard. Apache 2.0 license. Different names if you ask to Mistral or Nvidia, but same model, funny:  - Mistral: https://lnkd.in/dHAVi-hc  - Nvidia: https://lnkd.in/dmfNfMkv - 𝐆𝐏𝐓-𝟒𝐨 𝐦𝐢𝐧𝐢: OpenAI released a scaled down version of GPT-4o. Smarter than smaller closed source models including gpt-3.5-turbo (have a look at the pic, MMLU reported). 60% cheaper than gpt-3.5-turbo.Just $0.15/1M input and $0.60/1M. 128k context input. Last two characteristics make me think this is the 𝐛𝐞𝐬𝐭 𝐬𝐨𝐥𝐮𝐭𝐢𝐨𝐧 July 24 for 𝐑𝐀𝐆 𝐚𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧𝐬 (cheap, probably robust and multilingual with plenty of room for context). Text and image input for now, video and audio coming. - 𝐃𝐞𝐞𝐩𝐒𝐞𝐞𝐤-𝐕𝟐-𝟎𝟔𝟐𝟖: DeepSeek open-sourced their model becoming first open-source model in LMSYS Chatbot Arena.

The world of copilots + PCs. Microsoft, in an event yesterday, introduced the new PCs and the special Neural Processing Unit (NPU) for AI operations - Here is what you should know ▶ I have been saying this that the future is of Agents and it seems Microsoft is moving in this direction as well with the release of the copilot in your PC powered by powerful models like GPT-4o and different Small Language Models (SLMs) to run locally, directly on the device. Your own personal assistant that is able to understand audio, create or search images, translate in 40 languages and eventually (I assume) learn your daily usage habits to provide you best assistance ▶ Partnered with Qualcomm which is also a big step in a sense that Microsoft is also going to use Arm64 architecture and hence the relevant apps with more native experience based on this architecture other than Intel's ▶ By default, windows 11 is enabled with 'Pluton Security processor' for state of the art security in these Copilot+PC machines ▶ Also partnered with Adobe for several of their apps ▶ Recall feature -> Access everything you've seen or done on your PC like having photographic memory, with information organized by unique associations for quick, intuitive retrieval ▶ Seamless image creation and editing using Image Creator Worth checking the videos on this blog post -> https://lnkd.in/dyEGSppA _____________________________________ ✅ Follow me Naqqash Abbassi for regular updates and insights on Generative AI and my journey as a CTO in AI product development

Microsoft's #1-Bit #LLMs: A Game Changer for AI Efficiency. Excited to discuss Microsoft's groundbreaking research on 1-bit Large Language Models (LLMs)! These new models utilize ternary values (-1, 0, 1) instead of the traditional 16-bit floats, leading to some incredible benefits: 2.7x Faster Processing 3.5x Less GPU Memory Usage ⚡ 71x Reduction in Energy Consumption But here's the kicker: Bitnet, a 1-bit LLM, performs on par or even outperforms traditional models like LLaMA 3B! This suggests significant efficiency gains without sacrificing accuracy. What does this mean for the future? Wider deployment of LLMs on mobile phones and edge devices due to their smaller footprint. Increased accessibility of LLMs for a broader range of users and organizations. Read more about the same: https://lnkd.in/gDNQa3xj Development of specialized hardware for even greater efficiency. This is a major leap forward for LLMs, and I'm excited to see where it goes! #AI #MachineLearning #Microsoft #LLMs #Efficiency #FutureofTech

NVIDIA AI Workbench is a great tool if you want to chat with many open source LLMs from Meta, Mistral, Microsoft, Google, IBM etc. By following this video you can setup NVIDIA AI Workbench tool within 30 minutes and no GPU required. It also has a feature to upload PDFs so you can chat with your own data in a RAG manner. I found this tool very useful to compare responses from various open source large language models. https://lnkd.in/gv6HcjVV

𝐆𝐨𝐨𝐠𝐥𝐞’𝐬 𝐧𝐞𝐰 𝐬𝐦𝐚𝐥𝐥 𝐀𝐈 𝐦𝐨𝐝𝐞𝐥 𝐆𝐞𝐦𝐦𝐚 𝟐 𝟐𝐁 𝐬𝐮𝐫𝐩𝐚𝐬𝐬𝐞𝐬 𝐎𝐩𝐞𝐧𝐀𝐈’𝐬 𝐆𝐏𝐓-𝟑.𝟓 🤯 Google has introduced Gemma 2 2B, a compact yet powerful AI model that outperforms larger models like GPT-3.5 and Mistral 8x7B on key benchmarks. This lightweight model has achieved MMLU and MBPP scores 10% higher than its predecessor, showcasing exceptional conversational AI capabilities. Gemma 2 2B's standout features include its ability to run efficiently on a wide range of hardware, from edge devices and laptops to robust cloud deployments. It's optimized with the NVIDIA TensorRT-LLM library and seamlessly integrates with platforms like Keras, JAX, Hugging Face, NVIDIA NeMo, Ollama, Gemma.cpp, and soon on MediaPipe. With 2.6B parameters, trained on a massive 2 trillion token dataset, Gemma 2 2B matches the performance of significantly larger models in Chatbot Arena. Its open-source nature allows developers to download and integrate the model, paving the way for safer and more innovative AI applications. 𝐇𝐨𝐰 𝐝𝐨 𝐲𝐨𝐮 𝐭𝐡𝐢𝐧𝐤 𝐥𝐢𝐠𝐡𝐭𝐰𝐞𝐢𝐠𝐡𝐭 𝐦𝐨𝐝𝐞𝐥𝐬 𝐥𝐢𝐤𝐞 𝐆𝐞𝐦𝐦𝐚 𝟐 𝟐𝐁 𝐰𝐢𝐥𝐥 𝐢𝐦𝐩𝐚𝐜𝐭 𝐭𝐡𝐞 𝐟𝐮𝐭𝐮𝐫𝐞 𝐨𝐟 𝐀𝐈 𝐝𝐞𝐯𝐞𝐥𝐨𝐩𝐦𝐞𝐧𝐭? Discover more about Gemma 2 2B and its potential to revolutionize AI development on Google's official announcement page. (Link in the first comment below ⬇️) Follow AI Toolhouse - AI Tools Catalogue for more 😃 #google #gemma2 #gemini #api #code #ai #aitools

NVIDIA article on the emergence of Large Language Model (LLM) Agents and their transformative potential in AI development. This piece provides a comprehensive introduction to LLM Agents, diving into their capabilities, applications, and how they're set to revolutionize various industries. Whether you're an AI enthusiast, a developer, or someone curious about the future of technology, this article is a must-read. It brilliantly captures the essence of LLMs and their integration into practical solutions. Check it out here for a deep dive: Introduction to LLM Agents https://lnkd.in/dZMC9qej

Looking for AI/ML engineers building/deploying LLM apps in enterprise setting. While everyone talks about GPU cost and availability there're other important obstacles to consider. Ping me if you are up for a quick chat to share your experience. #genai #llm #machinelearning

🚀 Microsoft Build 2024: Unveiling the Future of AI and Computing 🚀 Just wrapped up an epic week at Microsoft Build 2024! The event showcased some groundbreaking updates that are set to transform the tech landscape as we know it. Here are the top 8 announcements that caught my attention: 1. Copilot Studio: Build custom AI copilots to automate business processes effortlessly. 2. GitHub Copilot Extensions: Customize and integrate with services like Azure, Docker, and Sentry for seamless development workflows. 3. Copilot + PCs: High-performance, energy-efficient PCs powered by Neural Processing Units (NPUs) bringing unparalleled speed and intelligence. 4. Phi-3 Vision: A multimodal model supporting visual reasoning tasks, interpreting charts, tables, text, and image inputs. 5. GPT-4o: Now available in Azure AI Studio, this model handles text, vision, and audio, setting new standards for AI experiences. 6. Windows Copilot Runtime: Extending AI capabilities across Windows with AI-infused layers and developer tools. 7. Surface Pro and Surface Laptop: Enhanced performance and AI capabilities with the power of Snapdragon X. 8. Real-Time Intelligence in Microsoft Fabric: Enabling faster decision-making with real-time data analysis. Each of these innovations promises to enhance productivity, creativity, and efficiency in ways we've only dreamed of. Excited to see how these updates will reshape our industry! Check out the event recap and videos to see these announcements in action. 🌐💡 #MicrosoftBuild #AI #TechInnovation #FutureOfTech #Microsoft #AIRevolution #Microsoft #MicrosoftBuild2024 #ChatGPT #GPT4o #OpenAI

Google Gemma going to be new llama2 Google Gemma 7B model is 17% better than llama2-13B , It’s pretty 🤩 good 👍 though 🌐 Global Launch: Gemma, a family of lightweight, state-of-the-art open models, is now available worldwide, with model weights for Gemma 2B and Gemma 7B, offering pre-trained and instruction-tuned variants. 🛠️ Tools for Innovation: Alongside Gemma, Google releases a Responsible Generative AI Toolkit and toolchains for JAX, PyTorch, and TensorFlow, facilitating developer innovation and responsible AI development. 💻 Accessible AI: Gemma models are optimized for performance across NVIDIA GPUs and Google Cloud TPUs, ensuring they can run efficiently on laptops, workstations, and cloud platforms. 🔒 Safety First: Designed with Google’s AI Principles in mind, Gemma models are rigorously tested and fine-tuned for safe and responsible outputs, supported by a Responsible Generative AI Toolkit for developers. 🔄 Optimization and Compatibility: Gemma models are fine-tunable across various frameworks and compatible with a wide range of devices and hardware, ensuring broad accessibility and industry-leading performance. 🚀 Free Resources: Google offers free access to Gemma through Kaggle and Colab For details dive in to below article #opensource #generatieveai #llm #largelanguagemodels #nlp #machinelearning #ml #ai #artificialintelligence #naturallanguageprocessing #datascience #googleai #responsibleai #toolkit https://lnkd.in/gg9tPEmF Model card - https://lnkd.in/gUuZxhgM Kick starter - https://lnkd.in/g8rGu9gn