Muhammad Firmansyah Kasim mfkasim1

## nvidia-checking-commands.md

      
              1 file
            
          
              0 forks
            
          
              0 comments
            
          
              0 stars
            
          
                mfkasim1
                / nvidia-checking-commands.md
            
            
              Last active
              August 3, 2020 09:47
            
              
                NVIDIA GPU checking commands
              
          
    Check available NVIDIA

lspci | grep -i nvidia

Check NVIDIA driver version

cat /proc/driver/nvidia/version

Display log for drivers


## jacobian.py
def bjac(gy, y):
    """
    Get the batched Jacobian matrix for vector-valued output, gy, with respect to the input, y.
    The Jacobian output will take shape of (nbatch, *gy.shape[1:], *y.shape[1:])
    """
    # reshape gy to be batched 1D vector
    ggy = gy.view(gy.shape[0], -1)
    outshape = (gy.shape[0], *gy.shape[1:], *y.shape[1:])

    nbatch, nfout = ggy.shape

## eig.py
import torch

class eig(torch.autograd.Function):
    @staticmethod
    def forward(ctx, A):
        # normalize the shape to be batched
        Ashape = A.shape
        if A.ndim == 2:
            A = A.unsqueeze(0)
        elif A.ndim > 3:

## guide.md

      
              1 file
            
          
              0 forks
            
          
              0 comments
            
          
              0 stars
            
          
                mfkasim1
                / guide.md
            
            
              Last active
              July 26, 2020 10:58
            
              
                Step by step in starting a deep learning project (pytorch)
              
          
    Supervised learning


Create the Dataset object and test if it can load the data correctly
Create the transformation (in DataLoader) for the Dataset
Create a model and test if the DataLoader can feed the data to the model
Create the training procedure

Tips

Datasets


## python_numpy_tricks.py
# plot histogram with logscale in the x-axis
import numpy as np
import matplotlib.pyplot as plt
def plot_loghist(x, bins=30, **hist_kwargs):
    hist, bins = np.histogram(x, bins=bins)
    logbins = np.logspace(np.log10(bins[0]),np.log10(bins[-1]),len(bins))
    plt.hist(x, bins=logbins, **hist_kwargs)
    plt.xscale('log')

# silencing loud third-party library

## debug_multilevel_grad.py
"""
Gradph object can be used to debug multi-level autograd for anomaly detection.
To see an example on how to use this, see the bottom of this script.
"""

import torch

class GradphNode(object):
    def __init__(self, gfn, parent_node):
        self._gfn = gfn

## useful_git_commands.md

      
              1 file
            
          
              0 forks
            
          
              0 comments
            
          
              0 stars
            
          
                mfkasim1
                / useful_git_commands.md
            
            
              Created
              December 12, 2022 14:41
            
              
                Useful git commands
              
          
    These are the git commands I usually use (for self-reminder), sorted from the most-used ones:

git status: to see the states
git log --oneline: to see which commit I'm at
git add <filename(s)>: to stage a file(s)
git commit -m "<Some message>": to make a commit
git push: push the latest update to the remote
git pull: pull the latest states from the remote
git clone <repo_url>: clone a new repository
git diff : to see which lines in the file that haven't been staged


## div-avx.cpp
// g++ -O2 div-avx.cc -I/home/muhammad/libs/sleef/include/ -mavx -L/home/muhammad/libs/sleef/lib/ -lsleef

#include <immintrin.h>
#include <sleef.h>
#include <chrono>
#include <iostream>
#include <iomanip>
#include <cmath>
#include <complex>
	def bjac(gy, y):
	"""
	Get the batched Jacobian matrix for vector-valued output, gy, with respect to the input, y.
	The Jacobian output will take shape of (nbatch, gy.shape[1:], y.shape[1:])
	"""
	# reshape gy to be batched 1D vector
	ggy = gy.view(gy.shape[0], -1)
	outshape = (gy.shape[0], gy.shape[1:], y.shape[1:])

	nbatch, nfout = ggy.shape
	import torch

	class eig(torch.autograd.Function):
	@staticmethod
	def forward(ctx, A):
	# normalize the shape to be batched
	Ashape = A.shape
	if A.ndim == 2:
	A = A.unsqueeze(0)
	elif A.ndim > 3:
	# plot histogram with logscale in the x-axis
	import numpy as np
	import matplotlib.pyplot as plt
	def plot_loghist(x, bins=30, **hist_kwargs):
	hist, bins = np.histogram(x, bins=bins)
	logbins = np.logspace(np.log10(bins[0]),np.log10(bins[-1]),len(bins))
	plt.hist(x, bins=logbins, **hist_kwargs)
	plt.xscale('log')

	# silencing loud third-party library
	"""
	Gradph object can be used to debug multi-level autograd for anomaly detection.
	To see an example on how to use this, see the bottom of this script.
	"""

	import torch

	class GradphNode(object):
	def __init__(self, gfn, parent_node):
	self._gfn = gfn
	// g++ -O2 div-avx.cc -I/home/muhammad/libs/sleef/include/ -mavx -L/home/muhammad/libs/sleef/lib/ -lsleef

	#include <immintrin.h>
	#include <sleef.h>
	#include <chrono>
	#include <iostream>
	#include <iomanip>
	#include <cmath>
	#include <complex>