跳至主要內容

博客主页
项目
技术
友情链接

Techniques

agicy小于 1 分钟

目录

专题：LeetGPU - 从零手写 CUDA 算子
- Challenges
  1D Convolution 2D Convolution 2D Jacobi Stencil 2D Max Pooling 2D Subarray Sum 3D Convolution 3D Subarray Sum All-Pairs Shortest Paths Attention with Linear Biases (ALiBi)Batch Normalization Batched Matrix Multiplication BFS Shortest Path Categorical Cross Entropy Loss Causal Self-Attention Color Inversion Count 2D Array Element Count 3D Array Element Count Array Element Dot Product Fast Fourier Transform (FFT)FP16 Batched Matrix Multiplication FP16 Dot Product Gaussian Blur Gaussian Error Gated Linear Unit General Matrix Multiplication (GEMM)GPT-2 Transformer Block Histogramming INT8 Quantized MatMul Interleave Arrays K-Means Clustering Leaky ReLU Linear Self-Attention Logistic Regression Matrix Addition Matrix Copy Matrix Multiplication Matrix Power Matrix Transpose Max Subarray Sum Mean Squared Error Merge Sorted Arrays MoE Top-K Gating Monte Carlo Integration Multi-Agent Simulation (Boids)Multi-Head Attention Nearest Neighbor Ordinary Least Squares Parallel Merge Prefix Sum Radix Sort Rainbow Table Reduction ReLU Reverse Array RGB to Grayscale RMS Normalization Rotary Positional Embedding (RoPE)Sigmoid Activation Sigmoid Linear Unit (SiLU)Simple Inference Sliding Window Self-Attention Softmax Softmax Attention Sorting Sparse Matrix-Dense Matrix Multiplication Sparse Matrix-Vector Multiplication Stream Compaction Subarray Sum Swish-Gated Linear Unit Top K Selection Top-p Sampling (Nucleus)Value Clipping Weight Dequantization 向量加法
专题：Transformer 深度解析 (Transformer Insider)
巧用 Linux 网络机制，打通网际孤岛
深入理解 RPC 语义：从可能交付到精确一次
深入理解内存一致性：从原子操作到指令集架构
深度解析流量治理——从重试风暴到自适应限流算法
解析循环冗余校验码——能检能纠的强大能力
详解 Git 的三种 Merge Request 形式

此处应有页脚。

Copyright © 2026 agicy