Clarifai’s Post

"Kinara, a specialist in energy-efficient artificial intelligence at the edge, has unveiled its Ara-2 processor — claiming enough power to run large language models (LLMs) and other generative AI models on device, with up to eight times the performance of its predecessor. "With Ara-2 added to our family of processors, we can better provide customers with performance and cost options to meet their requirements," claims Kinara's chief executive officer, Ravi Annavajjhala, of the new part. "For example, Ara-1 is the right solution for smart cameras as well as edge AI appliances with 2-8 video streams, whereas Ara-2 is strongly suited for handling 16-32+ video streams fed into edge servers, as well as laptops, and even high-end cameras." https://lnkd.in/ead3EEkN

When looking at AI at the edge and embedded systems, we always tend to focus on the processor requirements. However, the processor is only one part of the hardware stack, and memory will be the real game changer for AI in embedded systems. https://bit.ly/3RxtSW4 #AI #Cadence

With the ROM-6881, Advantech is introducing a SMARC 2.1 Computer-on-Module (COM) with the Rockchip chipsets RK3588 and RK3588J, which have processors with eight cores. With up to 6 TOPS AI inference performance, it is suitable for complex tasks. It supports high-resolution videos with 8K@30fps video encoding and 8K@60fps encoding, which ensures high-quality display and resolution. https://lnkd.in/ds8CMKeN #embedded #embeddedsystems #computervision 

Tripple speed! LLMs on mobile aren't new, but lightning-fast Transformers are. ⚡ Large Language Models (LLMs) are the powerhouse behind many seemingly intelligent features on our phones. But let's face it, they can be painfully slow. The team at Transformer-Lite has set out to change that. Their goal: to make LLMs run at breakneck speeds on mobile GPUs. To achieve this, they've developed a suite of optimization techniques. From symbolic expressions for dynamic shape inference to a novel FP4 quantization method, they've left no stone unturned. The results are nothing to laugh at. Transformer-Lite achieves prefill and decoding speeds of 121 token/s and 14 token/s for the hefty ChatGLM2 6B model. For the more compact Gemma 2B, it hits a blazing 330 token/s and 30 token/s. It's pretty clear by now that AI will transform almost everything we run on our phones. The question is: at what speed? ↓ Liked this post? Consider subscribing to my substack to support my work by following the link under my name 💡

"The critical X-factor in NeuReality’s NR1 chip is its impressive performance-to-cost ratio. Compared to conventional CPU-centric systems, the NR1 chip showcased a 10-fold improvement in performance at equivalent costs during trials," reports The Jerusalem Post. Our complete and versatile solution with our novel NAPU (the NR1 chip) can work with a large variety of AI applications and pipelines - from intelligent virtual assistants and translation documents to sentimental analysis and fraud detection. #WeMakeAIEasy #NAPU #NR1 #aiapplications #aisoftware #financialservices #healthtech #comingsoon #thefutureishere

When work began on Kandou’s landmark signal conditioning solution, our primary goal was to use mathematical precision to enable the purest signal possible. Chord™ Signaling was the result, and it’s been at the centre of Kandou’s product designs since 2010. Today, lots of system designers are trying to find ways to maintain efficiency across PCIe links, and cope with the constraints on memory capacity and bandwidth created by the memory wall – particularly those attempting to unlock AI at scale. How do you jump the memory wall and give AI access to the memory reserves it needs? Because absolute precision was hardwired into Chord Signaling’s design from the beginning, it offers a powerful tool – with the most efficient signal to noise ratio in the world. By scaling memory channels up to 8x and 16x with no signal drop, a big part of the answer to unlocking AI is right here. Check out our latest blog to read more. https://lnkd.in/d-HfWjga

NVIDIA releases Nemotron 340B, an open LLM matching GPT-4 performance. 🚀 What's New? NVIDIA recently launched the Nemotron 340B, a comprehensive suite designed for synthetic data generation that enhances the development of large language models (LLMs). This release includes three specialized models: the Nemotron 340B Base, Instruct, and Reward, each tailored to optimize different stages of data generation and model training. 🔑 Key Features - Advanced Data Generation - Integration and Optimization 🏋♂️ Training and Customization Options - Customization Through NeMo - Model Alignment ♿ Accessibility and Licensing - Wide Accessibility - Open Model License Would you like to know what AI can and can't do for your business? #llm #chatbot #aiforbusiness

MediaTek's Kompanio 838 Chipset for Enhanced AI Chromebooks https://lnkd.in/gCqCJGuW #MediaTeksKompanio838Chipset #838ChipsetforEnhancedAIChromebooks #Kompanio838Chipset #MediaTek #AIChromebooks #AINews #AnalyticsInsight #AnalyticsInsightMagazine

Get your future-ready laptop today with Ryzen AI

As 2023 draws to a close, TechInsights predicts significant advancements in three key areas poised for substantial activity 👉 CPUs, AI, and processor-adjacent technologies.  https://bit.ly/3ROJyDQ   From refined small language models in data centers to low-power edge applications, 2024 is all about embracing the beauty of small.     Click here to read our analyst predictions for 2024. https://bit.ly/3ROJyDQ