SrivastavaKshitij

let's suppose there are 4 commits on top of the latest master that we want to squash

commit 1
commit 2
commit 3
commit 4
origin/master, origin/HEAD, master

1. rebase on master with SHA

SrivastavaKshitij

#!/usr/bin/env python
# coding: utf-8

# # [TVMConf 2020] BYOC Tutorial Demo
# 
# Cody Yu ([email protected]), Zhi Chen ([email protected]) from AWS AI.
# 
# 
# 
# This demo has two parts. In the first part, we use a simple Relay graph to walkthrough the BYOC workflow. In the second part, we build a SSD model with TensorRT BYOC integration to showcase a real world application.

SrivastavaKshitij

There are two ways of invoking pdb
python -m pdb <application>; Then set the breakpoint b <lineno> ; then run c
write import pdb; pdb.set_trace() where you want the debugger to stop and just run python <applicatio> ; pdb will stop at that line
common commands are
a.  w (where) to get stack trace
b. c (continue) - it will run the app until the next breakpoint
c. n (next) - will execute the current line and move to the next
d. s (step) - will step into the current line if its a function

SrivastavaKshitij

Steps to install torch tvm inside a container

Pull nvidia cuda ngc container

docker pull nvcr.io/nvidia/cuda:10.0-cudnn7-devel-ubuntu18.04

Start docker container

docker run -e NVIDIA_VISIBLE_DEVICES=0 --gpus 0 -it --shm-size=1g --ulimit memlock=-1  --rm  -v $PWD:/workspace/work $docker_image   

SrivastavaKshitij

# Use C-o like we do in screen as action key
#unbind C-b
set -g prefix C-a
#bind-key C-b last-window
bind-key a send-prefix
bind-key C-n next-window
bind-key C-p previous-window

# Terminal emulator window title
set -g set-titles on

SrivastavaKshitij

set tabstop=8
set expandtab
set softtabstop=4
set shiftwidth=4
filetype plugin indent on
set number
set backspace=indent,start
set hlsearch

SrivastavaKshitij

import torch 
from torch import nn
class autoencoder(nn.Module):
    def __init__(self,downsizing_factor=None,in_channels=1):
        self.downsize = downsizing_factor
        self.in_channels = in_channels
        super(autoencoder,self).__init__()
        
        conv_modules=[]
        self.in_channels = self.in_channels

SrivastavaKshitij

import torch 
from torch import nn
class autoencoder(nn.Module):
    def __init__(self,downsizing_factor=None,in_channels=1):
        self.downsize = downsizing_factor
        self.in_channels = in_channels
        super(autoencoder,self).__init__()
        
        conv_modules=[]
        self.in_channels = self.in_channels

SrivastavaKshitij

import torch
def pf1(output,target,metric=None):
    d = output.data
    t = target.data
    TP = torch.nonzero(d*t).size(0)
    TN = torch.nonzero((d - 1) * (t - 1)).size(0)
    FP = torch.nonzero(d * (t - 1)).size(0)
    FN = torch.nonzero((d - 1) * t).size(0)
    precision = TP / (TP + FP)
    recall = TP / (TP + FN)

SrivastavaKshitij

import numpy as np
def pf(output,target,metric=None):
    TP = np.count_nonzero(data*target)
    TN = np.count_nonzero((data - 1) * (target - 1))
    FP = np.count_nonzero(data * (target - 1))
    FN = np.count_nonzero((data - 1) * target)
    precision = TP / (TP + FP)
    recall = TP / (TP + FN)
    F1 = 2 * precision * recall / (precision + recall)
    accuracy = (TP+TN)/(TP+TN+FP+FN)