ALLPCB
What is an AI Edge Compute Box?
Comprehensive overview of an AI edge compute box: definition, operation, features, security, and industry use cases for real-time edge computing and computer vision.
Why GPUs Are Essential for AI Training
Technical overview of why GPUs outperform CPUs for deep learning training: neural networks' matrix operations, parallelism, GPU architecture and GPGPU benefits.
NVIDIA AI Workbench Simplifies AI Development
NVIDIA AI Workbench simplifies AI development with tools for RAG apps, GPU setups, and model customization across systems.
Deepseek R1 Offline Deployment Guide
Technical guide to installing RKLLM-Toolkit and converting/deploying the DeepSeek-R1 LLM on EASY-EAI-Orin-Nano (RK3576), covering env setup, conversion, and on-device inference.
Memory Challenges in AI and Machine Learning Compute
Overview of AI memory demands and new technologies: capacity, bandwidth, latency, power, reliability, and adoption challenges for future AI systems.
Sensors Enable AI to Perceive the World
Edge AI and vision sensors overview: embedded ML platforms, real-time image processing, depth sensing, and on-device inference for industrial, smart city, and IoT applications.
Beginner's Guide to Training Large Language Models
Technical guide to scaling LLM training: analyzes memory usage, gradient accumulation, ZeRO, and tensor/data parallelism to improve throughput and GPU utilization.
Length Extrapolation with Positional Encoding
Survey of length extrapolation approaches analyzing positional encoding methods and their effects on extending model context windows for long-sequence inference.
TVS Diode Selection for Surge and ESD Protection in Robots
Technical overview of TVS diode protection for robotic systems: BMS short-circuit, motor drive and DC-DC surge schemes, plus signal interface ESD and EMC considerations.
Building a Multilingual Open-Ended QA Dataset
OMGEval presents an open-source multilingual open-ended QA benchmark (804 Chinese prompts) localized from AlpacaEval, using Text-Davinci-003 baseline and GPT-4 evaluation.
RDK X3 Tennis Ball Collector Robot
Autonomous tennis-ball collection and serving robot using YOLOv5 detection and SLAM/AMCL localization, with optimized path planning and hierarchical wheel control.
Deploy DeepSeek-R1 on RK3588 Development Board — Part 1
Technical guide to porting DeepSeek-R1 onto RK3588-based edge hardware: model conversion, cross-compilation, board deployment, and measured CPU, memory, and NPU performance.
MambaQuant: Post-Training Quantization for Mamba Models
MambaQuant: PTQ for Mamba models using KLT-enhanced and smoothed fused rotations to enable high-accuracy W8A8/W4A8 post-training quantization with <1% loss.
Understanding Generative Adversarial Networks (GANs)
Explains how generative adversarial networks (GANs) work and shows step-by-step training of a PyTorch GAN on MNIST, including network design and training loop.
In-Memory Compute Chips for ML Inference
System-level evaluation of compute-in-memory (CiM) for accelerating GEMM in ML inference: compares analog vs digital CiM, cache-level integration, and optimal dataflows.
AI Chip Ecosystem: Analysis and Future Outlook
Analysis of the AI chips ecosystem: technical, commercial, and strategic factors shaping developer affinity, multi?generation platform continuity, and ecosystem influence.
Random Forest: What It Is and How It Works
Technical overview of random forest algorithms, bagging, bias-variance tradeoffs, and GPU-accelerated implementations (RAPIDS) for faster model training.
What Is an RNN (Recurrent Neural Network)?
Technical overview of RNNs and LSTM architectures, how they model sequential data, application areas like signal and text processing, and MATLAB-based implementation.
Key Factors in Designing an AI Chip
Review of neural network quantization and numeric formats, covering floating vs integer, block floating point, logarithmic systems, and inference vs training trade-offs.
LLM Inference Optimization for NLP Applications
System-level overview of LLM inference optimization, detailing techniques and tradeoffs to improve throughput for Transformer-based large language models.