anthonyrisinger

package main

import (
	"bufio"
	"flag"
	"fmt"
	"io"
	"math/rand"
	"os"
	"slices"

tjbcg

#!/usr/bin/env bash

set -e

COLOR='\033[1;32m'
NC='\033[0m' # No Color

log() {
  echo -e "${COLOR}#" $@ "${NC}"
}

edwardmyung

Graphite.dev

• Site: https://graphite.dev
• Docs: https://docs.graphite.dev/

Why?

• Stacked PRs allow you to easily work like so:
  • Work on branch ACQ-000-part-A (which is based on master)
  • Work on branch ACQ-000-part-B (which is based on part-A)

y0ast

# Now available here: https://github.com/y0ast/pytorch-snippets/tree/main/subset_of_imagenet

y0ast

# Now available here: https://github.com/y0ast/pytorch-snippets/tree/main/minimal_cifar

sbarratt

def get_jacobian(net, x, noutputs):
    x = x.squeeze()
    n = x.size()[0]
    x = x.repeat(noutputs, 1)
    x.requires_grad_(True)
    y = net(x)
    y.backward(torch.eye(noutputs))
    return x.grad.data

apaszke

import torch                                                                                          
                                                                                                      
def jacobian(y, x, create_graph=False):                                                               
    jac = []                                                                                          
    flat_y = y.reshape(-1)                                                                            
    grad_y = torch.zeros_like(flat_y)                                                                 
    for i in range(len(flat_y)):                                                                      
        grad_y[i] = 1.                                                                                
        grad_x, = torch.autograd.grad(flat_y, x, grad_y, retain_graph=True, create_graph=create_graph)
        jac.append(grad_x.reshape(x.shape))                                                      

talkingmoose

(*

----------------------------------------------------------------------------------

ABOUT THIS SCRIPT

Written by:William Smith
Professional Services Engineer
Jamf
[email protected]

apaszke

def Rop(y, x, v):
  """Computes an Rop.
  
  Arguments:
    y (Variable): output of differentiated function
    x (Variable): differentiated input
    v (Variable): vector to be multiplied with Jacobian from the right
  """
  w = torch.ones_like(y, requires_grad=True)
  return torch.autograd.grad(torch.autograd.grad(y, x, w), w, v)

noopkat

⚠️ this is now stupidly out of date

Computers

• 13" Macbook Pro 3.3 GHz i7 (late 2016)
• Microsoft Surface Book (2016)

Peripherals

• Dell up3216q 32" monitor