jamesr66a

#ifndef CONV_TBC_OP_H
#define CONV_TBC_OP_H 

#include <ATen/ATen.h>
#include <caffe2/core/context.h>
#include <caffe2/core/operator.h>

namespace caffe2 {

using at::Half;

jamesr66a

#include "onnx.pb.h"

#include <algorithm>
#include <fstream>
#include <iostream>
#include <numeric>
#include <queue>
#include <tuple>
#include <unordered_map>
#include <unordered_set>

jamesr66a

jamesr66a

+-Module
  +-ClassDef(Sequence)
    +-FuncDef(__init__)
      +-Args
        +-Arg(self)
      +-Expr
        +-Call
          +-Attribute
            +-Call
              +-Name(super)

jamesr66a

    def test_script_fibb(self):
        cu = torch.jit._jit_script_compile('''
        def test_while(lim) -> (third):
            first = 1
            second = 1
            i = 1
            somenum = 5
            dontmutateme = 3
            third = 0 # TODO: python lexical scoping
            while i < lim:

jamesr66a

import torch
import torch.onnx

class MyCastModule(torch.nn.Module):
    def __init__(self):
        super().__init__()

    def forward(self, x):
        x = x.int()
        return torch.add(x, x)

jamesr66a

10:10:59 ~/onnx-fairseq/pytorch (fixpacktrace2) $ python test/test_jit.py
Traceback (most recent call last):
  File "test/test_jit.py", line 1, in <module>
    import torch
  File "/Users/jamesreed/onnx-fairseq/pytorch/torch/__init__.py", line 78, in <module>
    from torch._C import *
ImportError: dlopen(/Users/jamesreed/onnx-fairseq/pytorch/torch/_C.cpython-36m-darwin.so, 9): Library not loaded: @rpath/libATen.dylib
  Referenced from: /Users/jamesreed/onnx-fairseq/pytorch/torch/lib/libshm.dylib
 Reason: Incompatible library version: libshm.dylib requires version 1.0.0 or later, but libATen.dylib provides version 0.0.0

jamesr66a

scan_outputs_0 = ...
hidden_0 = ...

for t in range(timesteps):
  scan_outputs_0[t], hidden_0 = LSTM(inputs[t], hidden_0)
  
scan_outputs_1 = ...
hidden_1 = ... 
 
for t in range(timesteps):

jamesr66a

    def test_trace_size(self):
        def fn(x):
            return x.view(x.shape[1] * 2, x.size(0), 2)

        x = torch.randn(5, 2, 4, requires_grad=True)
        y = torch.randn(4, 8, 4)

        # Check that it behaves as expected
        traced_fn = torch.jit.trace(x)(fn)
        self.assertEqual(traced_fn(y), fn(y))

jamesr66a

import torch

class TracedForward(torch.jit.ScriptModule):
    @torch.jit.trace(torch.rand(5)):
    def forward(self, x):
        return torch.neg(x)