Markku Räsänen’s Post

We’re striving to make LLM quantization more reproducible to help the community, I think it’s a bit of an issue in the LLM space right now. We will soon add benchmarks to our model quants too (we want to get it just right first - it’s probably an even bigger issue) - but don’t expect a flood of models - our team is small and focused, just the best LLMs for their job (at the time of choosing it, this was the 2nd best ranking model for instruction following [IFEval] after Google’s proprietary Gemini Flash 002 based on livebench.ai and internal results). Big props to NeuralMagic for their open tooling and the ArceeAI team behind creating the great open source distill. It’s doing some magic in our AI systems :)

🪄 Wizard LM May Be The New SOTA Open-Source LLM 🪄 Sure, there are many new LLMs, but this one deserves a mention as this may be the top open-source model at the moment. WizardLM-2 8x22B may be in the top 5-6 LLMs, only behind Claude and GPT-4 The MT-Bench score of 9.12 is very impressive!! Every day, the gap between open and closed source narrows!! 👉 For Implementation of LLM Applications and Consultancy projects, feel free to DM us We would love to hear your thoughts! Stay in the loop with the updates by following Antar AI. Generative AI Muhammad Rizwan Munawar Piotr Skalski Roboflow Roc4Tech Arnaud Bastide Alessandro Ferrari #llm #rag #wizardlm #llmapps #sota

𝙃𝙤𝙬 𝙘𝙖𝙣 𝙮𝙤𝙪 "𝙆𝙚𝙚𝙥 𝙋𝙖𝙘𝙚" 𝙬𝙞𝙩𝙝 𝘼𝙄? It's not to late to catch up! George shares a true wealth of information in his ongoing series linked below.

As part of my personal AI / GPT education, I share my weekly reading (and insights) EVERY Sunday morning @ 9AM EST for approximately 30 minutes The link to all 94 AI webinars is below The challenge I see (and experience) - is with the deluge of AI content that crosses our feeds on a DAILY basis 1. What are the key AI trends you should focus on 2. What are the capabilities AI is enabling - and what skills do you need 3. How can you "think" in AI terms 4. How can you help your company in its AI efforts For me, it has been a fantastic journey to date - I have significantly increased by "AI Muscle Memory" just thru the reading and preparation. If you find value in the content - all I ask is that you share it with your networks !! #keeppace #AI #GPT #ChatGPT https://lnkd.in/eH4t9R7p

Published my first Lightning AI studio template on "Post-Training Quantization to GPTQ format and Evaluation". Now, you can easily try the code by just duplicating the template without any setup hassle. Head over to the link: https://lnkd.in/gs4_qDun #lightningai #llm #quantizedllm #evaluation

How do we measure the capabilities of AI systems? Unfortunately, we don't really know how to do this rigorously. In this COLM 2024 position paper, we introduce the idea of 𝐌𝐨𝐝𝐞𝐥 𝐌𝐞𝐭𝐫𝐨𝐥𝐨𝐠𝐲, organizing a distinct subfield on the measurement science of language model capabilities.

"Is O1 really worth the premium? 🤯 We battled it against the more affordable Sonnet 3.5 in a head-to-head battle of AI-powered algorithm generation and Reasoning . The results are surely not what you expect to be . Watch our latest video to see which model reigns supreme in speed, accuracy, and algorithm design. Plus, we've got a challenge for you: can you solve the puzzles these AIs couldn't? 🤔 #AI #ArtificialIntelligence #MachineLearning #DeepLearning #DataScience #NeuralNetworks #AIethics #AIresearch #AIdevelopment #AILeadership #GenerativeAI #LLMs #ChatGPT #AIethics #BiasinAI #AIforGood #AIandJobs #AIandSociety

In case you are interested in catching up with all the recent developments in the world of LLMs and AI, I have published a summary of exciting announcements, papers, tools, and educational resources. Watch at 2X here: https://lnkd.in/eEw9K-zR

Long awaited and now finally here – Stable Diffusion 3. Quick recap: Stable Diffusion is a series of text-to-image generative AIs that had its first release in 2022. It’s the most popular open-source image generation AI. JvM STABLES PRESENTS: LESSON LAIRNED! According to Stability AI, the developers of Stable Diffusion, SD3 is their „most capable text-to-image model with greatly improved performance in multi-subject prompts, image quality, and spelling abilities.“ Strong words. Let’s put that to the test. Check out our slide show and see for yourself! 📸👁️ #genAI #JvM #lessonlairned #slideshow Jung von Matt CREATORS #StableDiffusion3 #StabilityAI

One of the most exciting developments our icetana GPT! Check out the quick demo below. #icetanaAI #GPT #KSA #UAE #Qatar