iacolippo

import os

import ray
from ray import tune
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.utils.data import DataLoader, TensorDataset

iacolippo

import os

import ray
from ray import tune
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.utils.data import DataLoader, TensorDataset

iacolippo

iacolippo

#include <torch/extension.h>
#include <cmath>
#include <iostream>
#include <vector>


at::Tensor ex_forward(
    at::Tensor input
    ) {
    auto n_samples = input.size(0);

iacolippo

from time import time

import torch
import torch.nn as nn
import torch.nn.functional as F


class ConvNetV0(nn.Module):
    def __init__(self):
        super(ConvNetV0, self).__init__()

iacolippo

from torchvision.datasets import MNIST
import torchvision.transforms as transforms

import torch
import torch.legacy.nn as lnn
import torch.legacy.optim as loptim

train_dataset = MNIST(root='./data', 
                      train=True, 
                      transform=transforms.ToTensor(),  

iacolippo

from __future__ import print_function, division
import math
import torch
from torch.autograd import Function, Variable
import torch.nn as nn
import torch.nn.functional as F

from ErrorFeedback import EF
from FunctionErrorFeedback import ErrorFeedbackFunction
from SequentialSG import SequentialSG

iacolippo

from FunctionErrorFeedback import ErrorFeedbackFunction
from torch.autograd import Function, Variable
import torch
import torch.nn as nn

class EF(nn.Module):
    
    def __init__(self, layer_dim, error_dim):
        super(EF, self).__init__()
        self.feedback = torch.Tensor(error_dim, layer_dim)

iacolippo

import torch
from torch.autograd import Function, Variable


class ErrorFeedbackFunction(Function):
    # the forward pass consists in copying the input to the output
    @staticmethod
    def forward(ctx, input, feedback):
        ctx.save_for_backward(input, feedback)
        return input

iacolippo

import torch.nn as nn

class SequentialSG(nn.Sequential):
    def accGradParameters(self, input, gradOutput, scale=1):
        currentGradOutput = gradOutput
        currentModule = self.modules[-1]
        
        for i in range(len(self.modules)-1, 0, -1):
            previousModule = self.modules[i]
            if currentModule.__class__.name == 'ErrorFeedback':