Rajeev Sharma’s Post

Zyphra is excited to release Zamba2-7B, a state-of-the-art small language model. At the 7B scale, we outperform the leading models of Mistral, Google’s Gemma and Meta’s Llama3 series in both quality and performance. We believe Zamba2-7B is the leading model for running on-device and on consumer GPUs as well as for many enterprise applications which require a powerful but compact and efficient model for natural-language tasks. #Zamba2 #LanguageModels #AIInnovation #NaturalLanguageProcessing #OnDeviceAI #EdgeAI #EnterpriseAI #MachineLearning #StateOfTheArt #LLMOptimization https://lnkd.in/g9Jy-DUz

Banana Pi BPI-F3 #SpacemiT K1 #AI demo: Four-way video simultaneous reasoning demonstration. AI fusion computing power and its software stack practice under RISC-V architecture https://lnkd.in/e_tKJxTY #riscv #sbc #bananapi #raspberrypi #opensource #embedded

Banana Pi BPI-F3 #SpacemiT K1 #AI demo: Four-way video simultaneous reasoning demonstration. AI fusion computing power and its software stack practice under RISC-V architecture https://lnkd.in/e_tKJxTY #riscv #sbc #bananapi #raspberrypi #opensource #embedded

Dataflow is higher performance AND more energy efficient than legacy GPU architecture. That about sums up SambaNova Systems’ approach to sustainable AI computing. we are also enabling fast and simple integration with 3rd parties such as Gradio. https://lnkd.in/gwTYEtw9 #ai #genai #inference #efficiency #sustainability #llm

Proposed a new model architecture called MobileNets based on depthwise separable convolutions. We investigated some of the important design decisions leading to an efficient model. We then demonstrated how to build smaller and faster MobileNets using width multiplier and resolution multiplier by trading off a reasonable amount of accuracy to reduce size and latency. We then compared different MobileNets to popular models demonstrating superior size, speed and accuracy characteristics. We concluded by demonstrating MobileNet’s effectiveness when applied to a wide variety of tasks. As a next step to help adoption and exploration of MobileNets, we plan on releasing models in Tensor Flow.

👨💻 If you're working on an edge AI application, what could be the best way to reach peak performance and maximize the utilization of your hardware?   While traditional optimization techniques such as quantization and autotuning are helpful, you can actually gain the most significant boost in performance by selecting a model architecture that’s optimal for your hardware, dataset, and task.   ⚙️ Learn more > https://lnkd.in/geDRg_6r   #edgeai #computervision #deeplearning #neuralnetworks

Banana Pi BPI-F3 #SpacemiT K1 #AI demo: Four-way video simultaneous reasoning demonstration. AI fusion computing power and its software stack practice under RISC-V architecture https://lnkd.in/gRXyEvMx #riscv #sbc #bananapi #raspberrypi #opensource #embedded

AI21 Labs introduced Jamba, a production-scale Mamba implementation featuring a hybrid SSM-Transformer Mixture of Experts architecture. Key Points: - 52B parameters, with 12B active during generation - New architecture with Joint Attention and Mamba - Supports a context length of 256K tokens - Fits up to 140K context on a single GPU - Achieves 3X throughput on long contexts compared to Mixtral 8x7B - Released under Apache 2.0 license - Competes with open LLMs on Open LLM Leaderboard Benchmarks. Traditional Transformer architectures, although powerful, face challenges such as large memory requirements and slower inference with growing context lengths due to the quadratic bottleneck in the attention mechanism. Mamba architectures attempted to address these issues but fell short in model output quality. Jamba claims to leverage the strengths of both architectures by combining Transformer, Mamba, and mixture-of-experts (MoE) layers, aiming to optimize memory, throughput, and performance by efficiently utilizing 12 billion of its 52 billion parameters during inference. Model link: https://lnkd.in/gbkeNzKe Announcement Blog: https://lnkd.in/gWdVuGqK In-depth explanation on Mamba architecture: https://lnkd.in/gZr9x9gK #ai #huggingface #opensource #machinelearning #llm

Register for free online workshops by AMD Authorized Training Providers to learn how to: - Effectively utilize the AMD Versal™ AI Engine architecture and interfaces - Program the AI Engine, including kernels and graphs - Use the DSP Library, develop AI Engine designs with AMD Vitis™ Model Composer, and debug AI Engine applications Choose from these courses and more in the July training calendar here: https://lnkd.in/e5FcGuD5 #TogetherWeAdvance #AdaptiveComputing #Training #Versal #AI #Vitis

🖼️ As previously announced, Stability AI has today released the new 'Stable Diffusion 3.5 Medium' model. The model is available on HuggingFace and is said to have a slightly adapted architecture compared to the larger 'Stable Diffusion 3.5 Large' and 'Stable Diffusion 3.5 Large Turbo' variants. 👇Try the demo. #StabilityAI #StableDiffusion35