jamesr66a

class TracedModule(torch.nn.Module):
    def forward(self, x):
        x = x.type(torch.float32)
        return torch.floor(torch.sqrt(x) / 5.)

tm = torch.jit.trace(torch.rand(5))(TracedModule())

class ScriptModule(torch.jit.ScriptModule):
    @torch.jit.script_method
    def forward(self, x):

jamesr66a

In file included from torch/csrc/torch.cpp:2:
/Users/jamesreed/onnx-fairseq/pytorch/torch/csrc/autograd/generated/VariableType.h:41:36: error: 'storageWithAllocator' marked 'override' but does not override any member functions
  virtual std::unique_ptr<Storage> storageWithAllocator(int64_t size,at::Allocator* allocator) const override;
                                   ^
/Users/jamesreed/onnx-fairseq/pytorch/torch/csrc/autograd/generated/VariableType.h:31:8: warning: abstract class is marked 'final' [-Wabstract-final-class]
struct VariableType final : public at::Type {
       ^
/Users/jamesreed/onnx-fairseq/pytorch/torch/lib/tmp_install/include/ATen/Type.h:90:36: note: unimplemented pure virtual method 'storageWithAllocator' in 'VariableType'
  virtual std::unique_ptr<Storage> storageWithAllocator(int64_t size,std::unique_ptr<Allocator> allocator) const = 0;
                                   ^

jamesr66a

    def test_call_python_fn_from_script_module(self):
        def python_fn(x):
            return torch.neg(x)

        class ScriptMod(torch.jit.ScriptModule):
            def __init__(self):
                super(ScriptMod, self).__init__()
                self.param = torch.nn.Parameter(torch.rand(4, 3))

            @torch.jit.script_method

jamesr66a

import torch

x = torch.randn(3, 3, requires_grad=True)
print(x)

min = float('NaN')
max = 0.0
y = torch.clamp(x, min, max)
print('y', y)
y.sum().backward()

jamesr66a

======================================================================
ERROR: test_python_call_annotation (__main__.TestScript)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "test/test_jit.py", line 1932, in test_python_call_annotation
    @torch.jit.script
  File "/Users/jamesreed/onnx-fairseq/pytorch/torch/jit/__init__.py", line 373, in script
    graph = _jit_script_compile(torch._C.TypedDef(ast, schema), rcb)
TypeError: _jit_script_compile(): incompatible function arguments. The followingargument types are supported:
    1. (arg0: torch._C.TypedDef, arg1: Callable[[str], function]) -> torch._C.Graph

jamesr66a

import torch

class ClampMod(torch.nn.Module):
    def forward(self, x):
        return x.clamp(max=3.14, min=4.13)

import io

f = io.BytesIO()

jamesr66a

commit bd1b4dbe99e6040ded6c4ab2a59798536a2d6a45
Author: James Reed <jamesreed@fb.com>
Date:   Mon Jul 23 13:59:34 2018 -0700

    Fix zipfile export

diff --git a/test/expect/TestScript.test_script_module_file_export.expect b/test/expect/TestScript.test_script_module_file_export.expect
new file mode 100644
index 000000000..1d7ce966b
--- /dev/null

jamesr66a

commit 96d6beb5d300da71c2e5eee0eec9a012480af8f0
Author: James Reed <jamesreed@fb.com>
Date:   Mon Jul 23 15:47:43 2018 -0700

    Bugfix for stateful module export

diff --git a/test/test_jit.py b/test/test_jit.py
index 9363c0954..39deb0f74 100644
--- a/test/test_jit.py
+++ b/test/test_jit.py

jamesr66a

import torch

@torch.jit.script
def foo(x, y):
    for i in range(100):
        if torch.fmod(_to_tensor(i), 3) == 0:
            y += x
        else:
            x += y
    return x, y

jamesr66a

diff --git a/torch/onnx/symbolic.py b/torch/onnx/symbolic.py
index 43fff76a1..ca5702837 100644
--- a/torch/onnx/symbolic.py
+++ b/torch/onnx/symbolic.py
@@ -710,6 +710,12 @@ def type_as(g, self, other):
         return g.op("ATen", self, other, operator_s="type_as")
 
 
+@parse_args('v', 'is', 'v', 'v', 'f', 'i')
+def layer_norm(g, self, normalized_shape, weight, bias, eps, cudnn_enable):