Linux User Management
User Management 用户 添加用户 useradd username -d home dir username 设置家目录 -g group name username 设置用户所属组 删除用户 userdel username -r 删除用户, 同时删除他的家目录 设置...
Soong
Linux File Directory
File Directory 路径 查看当前绝对路径 pwd 切换路径 cd path . 代表当前路径 .. 代表上一级 目录 创建目录 mkdir option dir name -p 创建多级目录 删除目录 rmdir dir name 删...
Linux Directory Structure
Linux Directory Structure linux中一切都是文件 根目录下的文件结构 指令 /bin 存放最常用的指令 /sbin 存放root权限指令 家目录 ...
Torch Train with Steps
使用steps forward def forward_batch(batch, model, loss_fn, device): """ forward一个batch return: metric """ X, y = batch X, y = X.to(device), y.to(device) y_hat = model(X) l = loss_f...
Torch Train with Mulit GPU
多GPU训练 1. nn.DataParallel devices = [torch.device(f'cuda:{i}') for i in range(torch.cuda.device_count())] net = nn.DataParallel(net, device_ids=devices) for epoch in range(num_epochs): for X, ...
Torch Train with Epoch
使用epoch train def train_epoch(model, trainer, loss_fn, dataloader, device): """ train一个epoch return: 平均loss """ model.train() loss_sum = 0.0 for X, y in dataloader: ...
Torch Tricks
torch tricks Set Seed import numpy as np import torch import random def set_seed(seed): np.random.seed(seed) random.seed(seed) torch.manual_seed(seed) if torch.cuda.is_available()...
Torch Modify Model
Modify Model torch model is a callable object, u can visit it as a list or tuple change model to a nn.Sequential model_seq = nn.Sequential(model.children()[:]) get any layer`s output create_...
Torch Custom learning rate scheduler
Learning rate scheduler 带warmup的learning rate随余弦函数变化的scheduler Warmup Phase lr从0线性增加init_lr Decay Phase lr随cos逐步降低到0 通过调整num_cycles决定经过cos的多少个周期 import math import torch...
Torch Custom Criterion
Criterion LabelSmoothedCrossEntropy class LabelSmoothedCrossEntropyCriterion(nn.Module): def __init__(self, smoothing, ignore_index=None, reduce=True): super().__init__() self...