Upgrade check scripts to pytorch 1.0

* dismiss deprecated Variable semantics
* use dtype and device keywords
* add possibility to benchmark with float64

Author: Clément Pinard

benchmark.py

@@ -17,6 +17,7 @@
 parser.add_argument('-r', '--runs', type=int, default=100)
 parser.add_argument('--scale', choices=['s', 'ms', 'us'], default='us')
 parser.add_argument('-c', '--cuda', action='store_true')
+parser.add_argument('-d', '--double', action='store_true')
 options = parser.parse_args()
 
 if options.example == 'py':
@@ -27,16 +28,16 @@
     from cuda.lltm import LLTM
     options.cuda = True
 
-X = torch.randn(options.batch_size, options.features)
-h = torch.randn(options.batch_size, options.state_size)
-C = torch.randn(options.batch_size, options.state_size)
-rnn = LLTM(options.features, options.state_size)
+device = torch.device("cuda") if options.cuda else torch.device("cpu")
+dtype = torch.float64 if options.double else torch.float32
 
-if options.cuda:
-    X = X.cuda()
-    h = h.cuda()
-    C = C.cuda()
-    rnn.cuda()
+kwargs = {'dtype': dtype,
+          'device': device,
+          'requires_grad': True}
+X = torch.randn(options.batch_size, options.features, **kwargs)
+h = torch.randn(options.batch_size, options.state_size, **kwargs)
+C = torch.randn(options.batch_size, options.state_size, **kwargs)
+rnn = LLTM(options.features, options.state_size).to(device, dtype)
 
 # Force CUDA initialization
 new_h, new_C = rnn(X, (h, C))

check.py

@@ -5,8 +5,6 @@
 import numpy as np
 import torch
 
-from torch.autograd import Variable
-
 import python.lltm_baseline
 import cpp.lltm
 
@@ -85,21 +83,23 @@ def check_backward(variables, with_cuda, verbose):
 
 if options.cuda:
     import cuda.lltm
-    options.cuda = True
-
-X = torch.randn(options.batch_size, options.features)
-h = torch.randn(options.batch_size, options.state_size)
-C = torch.randn(options.batch_size, options.state_size)
-W = torch.randn(3 * options.state_size, options.features + options.state_size)
-b = torch.randn(1, 3 * options.state_size)
+    device = torch.device("cuda")
+else:
+    device = torch.device("cpu")
+
+kwargs = {'dtype': torch.float64,
+          'device': device,
+          'requires_grad': True}
+X = torch.randn(options.batch_size,
+                options.features,
+                **kwargs)
+h = torch.randn(options.batch_size, options.state_size, **kwargs)
+C = torch.randn(options.batch_size, options.state_size, **kwargs)
+W = torch.randn(3 * options.state_size, options.features + options.state_size, **kwargs)
+b = torch.randn(1, 3 * options.state_size, **kwargs)
 
 variables = [X, W, b, h, C]
 
-for i, var in enumerate(variables):
-    if options.cuda:
-        var = var.cuda()
-    variables[i] = Variable(var.double(), requires_grad=True)
-
 if 'forward' in options.direction:
     check_forward(variables, options.cuda, options.verbose)
 

grad_check.py

@@ -3,8 +3,7 @@
 
 import argparse
 import torch
-
-from torch.autograd import Variable, gradcheck
+from torch.autograd import gradcheck
 
 parser = argparse.ArgumentParser()
 parser.add_argument('example', choices=['py', 'cpp', 'cuda'])
@@ -22,18 +21,20 @@
     from cuda.lltm import LLTMFunction
     options.cuda = True
 
-X = torch.randn(options.batch_size, options.features)
-h = torch.randn(options.batch_size, options.state_size)
-C = torch.randn(options.batch_size, options.state_size)
-W = torch.randn(3 * options.state_size, options.features + options.state_size)
-b = torch.randn(1, 3 * options.state_size)
+device = torch.device("cuda") if options.cuda else torch.device("cpu")
+
+kwargs = {'dtype': torch.float64,
+          'device': device,
+          'requires_grad': True}
+
+X = torch.randn(options.batch_size, options.features, **kwargs)
+h = torch.randn(options.batch_size, options.state_size, **kwargs)
+C = torch.randn(options.batch_size, options.state_size, **kwargs)
+W = torch.randn(3 * options.state_size, options.features + options.state_size, **kwargs)
+b = torch.randn(1, 3 * options.state_size, **kwargs)
 
 variables = [X, W, b, h, C]
 
-for i, var in enumerate(variables):
-    if options.cuda:
-        var = var.cuda()
-    variables[i] = Variable(var.double(), requires_grad=True)
 
 if gradcheck(LLTMFunction.apply, variables):
     print('Ok')