mirror of
https://github.com/saymrwulf/onnxruntime.git
synced 2026-07-10 17:37:14 +00:00
d521efd904
* refactor frontend * remove training python files from inferencing build * update according to reviewer's comments * merge pybind_state.cc * refactor pybind_state.cc * code clean up * missed a forward declaration in ort_pybind_state.cc * passed pytest * move training_session.py into a subfolder per reviewer's comment * add copyright Co-authored-by: liqun <liqun@OrtTrainingDev4.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
195 lines
8.5 KiB
Python
195 lines
8.5 KiB
Python
## This code is from https://github.com/pytorch/examples/blob/master/mnist/main.py
|
|
## with modification to do training using onnxruntime as backend on cuda device.
|
|
## A private PyTorch build from https://aiinfra.visualstudio.com/Lotus/_git/pytorch (ORTTraining branch) is needed to run the demo.
|
|
## To run the demo with ORT backend:
|
|
## python mnist_training.py --use-ort
|
|
## or
|
|
## python mnist_training.py --use-ort --use-ort-trainer
|
|
|
|
## When "--use-ort" is not given, it will run training with PyTorch as backend.
|
|
## Model testing is not complete.
|
|
|
|
from __future__ import print_function
|
|
import argparse
|
|
import torch
|
|
import torch.nn as nn
|
|
import torch.nn.functional as F
|
|
import torch.optim as optim
|
|
from torchvision import datasets, transforms
|
|
import numpy as np
|
|
|
|
# TODO: remove after ready for CV
|
|
# import sys
|
|
# sys.path.insert(0, '/bert_ort/liqun/onnxruntime/build/Linux/Debug/')
|
|
# import onnxruntime as ort
|
|
|
|
# sys.path.insert(0, '/bert_ort/liqun/onnxruntime/onnxruntime/python/')
|
|
# from ort_trainer import IODescription, ModelDescription, ORTTrainer, ORTModel
|
|
|
|
from onnxruntime.capi.ort_trainer import IODescription, ModelDescription, ORTTrainer, ORTModel
|
|
|
|
class NeuralNet(nn.Module):
|
|
def __init__(self, input_size, hidden_size, num_classes):
|
|
super(NeuralNet, self).__init__()
|
|
self.fc1 = nn.Linear(input_size, hidden_size)
|
|
self.relu = nn.ReLU()
|
|
self.fc2 = nn.Linear(hidden_size, num_classes)
|
|
|
|
def forward(self, x):
|
|
out = self.fc1(x)
|
|
out = self.relu(out)
|
|
out = self.fc2(out)
|
|
return out
|
|
|
|
def my_loss(x, target):
|
|
return F.nll_loss(F.log_softmax(x, dim=1), target)
|
|
|
|
def train_with_model(args, model, device, train_loader, optimizer, epoch):
|
|
model.train()
|
|
for batch_idx, (data, target) in enumerate(train_loader):
|
|
data, target = data.to(device), target.to(device)
|
|
data = data.reshape(data.shape[0], -1)
|
|
loss, pred = model.run(data, target)
|
|
if batch_idx % args.log_interval == 0:
|
|
optimizer.step()
|
|
optimizer.zero_grad()
|
|
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
|
|
epoch, batch_idx * len(data), len(train_loader.dataset),
|
|
100. * batch_idx / len(train_loader), loss.item()))
|
|
|
|
def test_with_model(args, model, device, test_loader, optimizer, epoch):
|
|
model.eval()
|
|
test_loss = 0
|
|
correct = 0
|
|
with torch.no_grad():
|
|
for data, target in test_loader:
|
|
data, target = data.to(device), target.to(device)
|
|
data = data.reshape(data.shape[0], -1)
|
|
|
|
pred = model.run(data, target, )
|
|
|
|
output = F.log_softmax(model.eval(data), dim=1)
|
|
test_loss += F.nll_loss(output, target, reduction='sum').item() # sum up batch loss
|
|
pred = output.argmax(dim=1, keepdim=True) # get the index of the max log-probability
|
|
correct += pred.eq(target.view_as(pred)).sum().item()
|
|
|
|
test_loss /= len(test_loader.dataset)
|
|
|
|
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
|
|
test_loss, correct, len(test_loader.dataset),
|
|
100. * correct / len(test_loader.dataset)))
|
|
|
|
def train_with_trainer(args, trainer, device, train_loader, epoch):
|
|
for batch_idx, (data, target) in enumerate(train_loader):
|
|
data, target = data.to(device), target.to(device)
|
|
data = data.reshape(data.shape[0], -1)
|
|
|
|
learning_rate = torch.tensor([args.lr])
|
|
loss = trainer.train_step((data, target, learning_rate))
|
|
|
|
if batch_idx % args.log_interval == 0:
|
|
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
|
|
epoch, batch_idx * len(data), len(train_loader.dataset),
|
|
100. * batch_idx / len(train_loader), loss.item()))
|
|
|
|
# TODO: comple this once ORT training can do evaluation.
|
|
def test_with_trainer(args, trainer, device, test_loader):
|
|
test_loss = 0
|
|
correct = 0
|
|
with torch.no_grad():
|
|
for data, target in test_loader:
|
|
data, target = data.to(device), target.to(device)
|
|
data = data.reshape(data.shape[0], -1)
|
|
output = F.log_softmax(trainer.eval_step(data, fetches=['probability']), dim=1)
|
|
test_loss += F.nll_loss(output, target, reduction='sum').item() # sum up batch loss
|
|
pred = output.argmax(dim=1, keepdim=True) # get the index of the max log-probability
|
|
correct += pred.eq(target.view_as(pred)).sum().item()
|
|
|
|
test_loss /= len(test_loader.dataset)
|
|
|
|
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
|
|
test_loss, correct, len(test_loader.dataset),
|
|
100. * correct / len(test_loader.dataset)))
|
|
|
|
def mnist_model_description():
|
|
input_desc = IODescription('input1', ['batch', 784], torch.float32)
|
|
label_desc = IODescription('label', ['batch', ], torch.int64, num_classes=10)
|
|
loss_desc = IODescription('loss', [], torch.float32)
|
|
probability_desc = IODescription('probability', ['batch', 10], torch.float32)
|
|
return ModelDescription([input_desc, label_desc], [loss_desc, probability_desc])
|
|
|
|
def main():
|
|
#Training settings
|
|
parser = argparse.ArgumentParser(description='PyTorch MNIST Example')
|
|
parser.add_argument('--batch-size', type=int, default=64, metavar='N',
|
|
help='input batch size for training (default: 64)')
|
|
parser.add_argument('--test-batch-size', type=int, default=1000, metavar='N',
|
|
help='input batch size for testing (default: 1000)')
|
|
parser.add_argument('--epochs', type=int, default=10, metavar='N',
|
|
help='number of epochs to train (default: 10)')
|
|
parser.add_argument('--lr', type=float, default=0.01, metavar='LR',
|
|
help='learning rate (default: 0.01)')
|
|
parser.add_argument('--momentum', type=float, default=0.5, metavar='M',
|
|
help='SGD momentum (default: 0.5)')
|
|
parser.add_argument('--no-cuda', action='store_true', default=False,
|
|
help='disables CUDA training')
|
|
parser.add_argument('--seed', type=int, default=1, metavar='S',
|
|
help='random seed (default: 1)')
|
|
parser.add_argument('--log-interval', type=int, default=10, metavar='N',
|
|
help='how many batches to wait before logging training status')
|
|
|
|
parser.add_argument('--save-model', action='store_true', default=False,
|
|
help='For Saving the current Model')
|
|
|
|
parser.add_argument('--use-ort', action='store_true', default=False,
|
|
help='to use onnxruntime as training backend')
|
|
|
|
parser.add_argument('--use-ort-trainer', action='store_true', default=False,
|
|
help='to use onnxruntime as training backend')
|
|
|
|
args = parser.parse_args()
|
|
use_cuda = not args.no_cuda and torch.cuda.is_available()
|
|
|
|
torch.manual_seed(args.seed)
|
|
|
|
device = torch.device("cuda" if use_cuda else "cpu")
|
|
|
|
kwargs = {'num_workers': 0, 'pin_memory': True}
|
|
train_loader = torch.utils.data.DataLoader(
|
|
datasets.MNIST('../data', train=True, download=True,
|
|
transform=transforms.Compose([
|
|
transforms.ToTensor(),
|
|
transforms.Normalize((0.1307,), (0.3081,))
|
|
])),
|
|
batch_size=args.batch_size, shuffle=True, **kwargs)
|
|
test_loader = torch.utils.data.DataLoader(
|
|
datasets.MNIST('../data', train=False, transform=transforms.Compose([
|
|
transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])),
|
|
batch_size=args.test_batch_size, shuffle=True, **kwargs)
|
|
|
|
input_size = 784
|
|
hidden_size = 500
|
|
num_classes = 10
|
|
model = NeuralNet(input_size, hidden_size, num_classes)
|
|
|
|
model_desc = mnist_model_description()
|
|
if args.use_ort_trainer:
|
|
# use log_interval as gradient accumulate steps
|
|
trainer = ORTTrainer(model, my_loss, model_desc, "SGDOptimizer", None, IODescription('Learning_Rate', [1,], torch.float32), device)
|
|
|
|
for epoch in range(1, args.epochs + 1):
|
|
train_with_trainer(args, trainer, device, train_loader, epoch)
|
|
import pdb
|
|
test_with_trainer(args, trainer, device, test_loader)
|
|
else:
|
|
model = ORTModel(model, my_loss, model_desc, device)
|
|
|
|
optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum)
|
|
|
|
for epoch in range(1, args.epochs + 1):
|
|
train_with_model(args, model, device, train_loader, optimizer, epoch)
|
|
# test(args, model, device, test_loader)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|