dhy2000

0.0.0.0/5
8.0.0.0/7
11.0.0.0/8
12.0.0.0/6
16.0.0.0/4
32.0.0.0/3
64.0.0.0/2
128.0.0.0/3
160.0.0.0/5
168.0.0.0/6

dhy2000

{
    "configurations": [
        {
            "name": "GDB Connect",
            "type": "cppdbg",
            "request": "launch",
            "program": "${workspaceFolder}/build/executable",
            "cwd": "${workspaceFolder}",
            "MIMode": "gdb",
            "miDebuggerServerAddress": "localhost:9999",

dhy2000

/**
 * Vector addition: C = A + B.
 *
 * This sample is a very basic sample that implements element by element
 * vector addition. It is the same as the sample illustrating Chapter 2
 * of the programming guide with some additions like error checking.
 */

#include <stdio.h>

dhy2000

FROM debian:bookworm

# setup apt
RUN sed -i 's#deb.debian.org#mirrors.ustc.edu.cn#g' /etc/apt/sources.list.d/debian.sources && \
    apt-get update && apt-get install -y apt-transport-https ca-certificates gpg && \
    sed -i 's#http#https#g' /etc/apt/sources.list.d/debian.sources && \
    apt-get update && apt-get upgrade -y  && apt-get install -y wget curl && apt-get clean

# install llvm
RUN wget --no-hsts -qO- https://apt.llvm.org/llvm-snapshot.gpg.key | tee /etc/apt/trusted.gpg.d/apt.llvm.org.asc && \

dhy2000

# Environment setup script for bash/zsh.
_ROOT=$(realpath $(dirname "${BASH_SOURCE[0]:-$0}"))

# --- ^_^ YOU CAN EDIT BELOW ^_^ --- #
_name="CUDA"
_components=(
  bin/nvcc
)
_add_path=(
  bin

dhy2000

Windows 11 Setup

跳过联网

Windows 11 家庭版初始化阶段，选择完键盘布局后可能进入强制连接无线网络并登录微软账号的环节（如果连接了网线需要先拔掉）。跳过无线联网的方法参考这篇CSDN博客，总共提到了三种方法，推荐采用第二种：

Shift+F10 或 Shift+Fn+F10 打开 cmd，输入 oobe\bypassnro.cmd 命令（如果 cmd 窗口不是当前焦点，无法输入命令，需按 Alt+Tab 切换窗口），等待系统重启后，来到联网这步选择“我没有 Internet 连接”，下一步选择“继续执行受限的设置”即可设置本地的用户名和密码，无需登录微软账户。

常规设置

dhy2000

## ignore known_hosts
StrictHostKeyChecking no
UserKnownHostsFile /dev/null

## keep alive
ServerAliveInterval 30
ServerAliveCountMax 100

Compression yes

dhy2000

// nvcc blas_bench.cu -o blas_bench -lcublas

/*
  Hipify
    hipify-clang blas_bench.cu -o blas_bench.hip

  Manually replace call of cublasGemmEx (with code in comment)

  Build with HIP
    hipcc -x hip blas_bench.hip -o blas_bench_hip -I /opt/rocm/include/hipblas -L /opt/rocm/lib -lhipblas

dhy2000

模型结构以及推理过程

Config

模型尺寸：

• 隐藏层大小 hidden_size （输入句子 embedding 之后的维度）
• 隐藏层数量 num_hidden_layers （即 Decoder Layer 的数量）
• 中间层大小 intermediate_size （MLP 层上投影和门投影之后的维度）

dhy2000

服务器开发环境建立

Zsh 配置

选用 oh-my-zsh 框架以及 powerlevel10k 主题。

主题下载

下载 oh-my-zsh 配置框架及实用插件，以及 powerlevel10k 主题：