devymex

import os, torch, torch.multiprocessing as mp
from transformers import AutoModelForCausalLM, LlamaConfig

CHECK_EQUAL_WITH_HF = '' # A pretrain directory eg. '/data/models/llama-2-hf/7b-chat'

def add_arguments(parser):
    group = parser.add_argument_group(title='Llama-2 HF saver.')
    group.add_argument('--megatron-path', type=str, default=None,
                       help='Base directory of megatron checkpoint')

devymex

#include <algorithm>
#include <cstring>
#include <fstream>
#include <iostream>

template<typename _Stream>
size_t GetFileSize(_Stream &stream) {
    auto nCurPos = stream.tellg();
    stream.seekg(0, std::ios::end);
    auto nFileSize = stream.tellg();

devymex

#! /bin/bash -e

export TOOLCHAIN_ROOT="/heop/.tool/toolchain/arm-ca9-linux-gnueabihf-6.5"
export PATH=$TOOLCHAIN_ROOT/bin:$PATH
export CPPFLAGS="--sysroot=$TOOLCHAIN_ROOT/arm-ca9-linux-gnueabihf/sysroot"
export AR=arm-linux-ar
export AS=arm-linux-as
export LD=arm-linux-ld
export RANLIB=arm-linux-ranlib
export CC=arm-linux-gcc

devymex

#!/bin/bash -e

export TOOLCHAIN_ROOT=/heop/.tool/toolchain/arm-ca9-linux-gnueabihf-6.5
export SYSROOT=$TOOLCHAIN_ROOT/arm-ca9-linux-gnueabihf/sysroot
export CC=$TOOLCHAIN_ROOT/bin/arm-linux-gcc
export CXX=$TOOLCHAIN_ROOT/bin/arm-linux-g++
export AR=$TOOLCHAIN_ROOT/bin/arm-linux-ar
export RANLIB=$TOOLCHAIN_ROOT/bin/arm-linux-ranlib

mkdir -p build

devymex

下载镜像：

docker pull <IMAGE_NAME> # https://hub.docker.com/r/nvidia/cuda

列出所有镜像

docker images

运行镜像（运行为一个容器）

devymex

#include "NvEncoder/NvEncoderCuda.h"
#include "NvEncoder/NvEncoderCLIOptions.h"
#include "NvEncoder/NvCodecUtils.h"
#include <cuda_runtime.h>

extern "C" {
#include <libavutil/opt.h>
#include <libavcodec/avcodec.h>
#include <libavformat/avformat.h>
}

devymex

/**
* Copyright (C) DeepGlint, Inc - All Rights Reserved
* Unauthorized copying of this file, via any medium is strictly prohibited
* Proprietary and confidential
*
* CTimer for Code Profile
*
* Written by Devymex <yumengwang@deepglint.com>, Jan. 2019
*/

devymex

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as checkpoint
import numpy as np
import math

def drop_path(x, drop_prob: float = 0., training: bool = False, scale_by_keep: bool = True):
    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
    This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,

devymex

#!/bin/bash -e

if [ $# -eq 0 ]; then
    echo "Please specify an name for the new container!"
    exit
fi
indices=""
for cid in $(docker container ls -a | sed 1,1d | awk '{print $1}'); do
    port=$(docker inspect --format='{{if .NetworkSettings.Ports}}{{(index (index .NetworkSettings.Ports "22/tcp") 0).HostPort}}{{else}}{{"null"}}{{end}}' $cid)
    indices=$indices" "$port

devymex

FROM nvidia/cuda:11.3.0-cudnn8-devel-ubuntu18.04
RUN mkdir -p /home/root
RUN apt update

RUN apt upgrade -y
RUN apt install -y apt-utils dialog
RUN apt install -y bash-completion locate vim tmux unzip htop
RUN apt install -y openssh-server # net-tools iputils-ping curl
RUN apt install -y build-essential pkg-config gdb git
RUN apt install -y libssl-dev libgoogle-glog-dev libgflags-dev