Ashish Patel 🇮🇳’s Post

Learning Series:Introduction LLM Quantization 1. Introduction to Weight Quantization : https://lnkd.in/dBEGS2BT 2. 4-bit LLM Quantization with GPTQ: https://lnkd.in/dnbzbXZq 3. Quantize Llama models with GGUF and llama.cpp: https://lnkd.in/d9_89prq 4. ExLlamaV2: The Fastest Library to Run LLMs:  https://lnkd.in/dWMCwFjk P.S. If you have similar resources please share in the comment

Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone by Microsoft We're thrilled to introduce phi-3-mini, a groundbreaking language model that fits right in your pocket! With 3.8 billion parameters trained on 3.3 trillion tokens, it rivals giants like GPT-3.5 in performance—all from your phone. 🔍 Our unique approach utilizes a mix of refined web data and synthetic data, enabling top-tier performance without the bulk. The phi-3-mini not only promises powerful AI capabilities on mobile devices but does so with an emphasis on safety and reliability, aligning with Microsoft’s responsible AI principles. 📊 Showcased results from benchmarks such as MMLU and HellaSwag confirm that phi-3-mini isn't just small, it's mighty, offering insights and interactions previously only possible on high-end servers. 🌐 Looking ahead, they're committed to enhancing multilingual support and further refining our models to handle a wider array of tasks more effectively. Stay tuned as we continue to push the boundaries of what's possible in AI technology, making it more accessible than ever. #AI #Innovation #Microsoft #LanguageModels

Introducing LLaVA-Llama-3-8B is released! XTuner team releases the new multi-modal models (LLaVA-Llama-3-8B and In the realm of multi-modal models, the XTuner team has taken a monumental leap forward with the launch of LLaVA-Llama-3-8B and its variant, LLaVA-Llama-3-8B-v1.1, powered by the groundbreaking Llama-3 LLM. These models aren't just a step up from their predecessors—they're a giant leap, smashing previous benchmarks and setting new standards of excellence. Why is this a game-changer? 🌟 The performance of these models doesn't just surpass the previous iteration; it obliterates it, redefining what we thought was possible in the process. And guess what? There's even more on the horizon with LLaVA-Llama-3-70B. For those looking to dive deeper, explore the models and their capabilities here: Model: https://lnkd.in/gC2YUrmB) https://lnkd.in/gGXZjyV8 Code: https://lnkd.in/gPYs7qgi ↓ Check out https://lnkd.in/gredE7Nu to get a weekly summary of the top models, repos and papers in AI. Read by 180,000+ engineers and researchers.

LLaVA-Llama-3-8B is released! XTuner team releases the new multi-modal models (LLaVA-Llama-3-8B and LLaVA-Llama-3-8B-v1.1) with Llama-3 LLM, achieving much better performance on various benchmarks. The performance evaluation substantially surpasses Llama-2. (LLaVA-Llama-3-70B is coming soon!) Model: https://lnkd.in/gC2YUrmB) https://lnkd.in/gGXZjyV8 Code: https://lnkd.in/gPYs7qgi

Exciting news in the LLM community! 🚀 The latest rankings are in, and it’s a close competition at the top. With GPT-4 variants showing strong performance, we see a notable entry with Llama-3-70b-Instruct securing a spot within the top ranks. This is more than a scoreboard—it’s a testament to the rapid advancements in LLMs language models. Organizations are pushing boundaries, as shown by the significant movement in positions. A key takeaway for anyone in tech or AI: stay adaptive, innovative, and ready to embrace the evolving landscape.