dhruvbird

import zipfile
bytecode_pkl = None
with zipfile.ZipFile('AddTensorsModelOptimized.ptl', 'r') as myzip:
  bytecode_pkl = myzip.open("AddTensorsModelOptimized/bytecode.pkl")
 
from torch.utils import show_pickle
show_pickle.DumpUnpickler.dump(bytecode_pkl, None)

(5,
 ('__torch__.___torch_mangle_12.AddtensorsModel.forward',

dhruvbird

graph(%self : __torch__.___torch_mangle_47.AddtensorsModel,
%x.1 : Tensor,
%y.1 : Tensor):
%3 : int = prim::Constant[value=1]()
%self.t1 : Tensor = prim::Constant[value= 0 5 10 [ CPUFloatType{3} ]]()
%5 : Tensor = aten::add(%self.t1, %x.1, %3) # <ipython-input-98-b27af2abc3c5>:8:8
%z.5 : Tensor = aten::add(%5, %y.1, %3) # <ipython-input-98-b27af2abc3c5>:8:8
return (%z.5)

dhruvbird

graph(%self : __torch__.___torch_mangle_41.AddtensorsModel,
%x.1 : Tensor,
%y.1 : Tensor):
%6 : int = prim::Constant[value=1]()
%z.1 : Tensor = prim::GetAttr[name="t1"](%self)
%7 : Tensor = aten::add(%z.1, %x.1, %6) # <ipython-input-98-b27af2abc3c5>:8:8
%z.5 : Tensor = aten::add(%7, %y.1, %6) # <ipython-input-98-b27af2abc3c5>:8:8
return (%z.5)

dhruvbird

graph(%self : __torch__.___torch_mangle_41.AddtensorsModel,
%x.1 : Tensor,
%y.1 : Tensor):
%5 : Tensor = prim::CallMethod[name="helper"](%self, %x.1, %y.1) # <ipython-input-98-b27af2abc3c5>:12:11
return (%5)

dhruvbird

#include <iostream>
#include <vector>
#include <torch/csrc/jit/mobile/import.h>
#include <torch/csrc/jit/mobile/module.h>

int main() {
  auto model = torch::jit::_load_for_mobile("AddTensorsModelOptimized.ptl"));

  std::vector<at::IValue> inputs, ret;
  inputs.push_back(at::zeros({3}));

dhruvbird

optimized_model._save_for_lite_interpreter("AddTensorsModelOptimized.ptl")

dhruvbird

from torch.utils.mobile_optimizer import optimize_for_mobile
optimized_model = optimize_for_mobile(scripted)

dhruvbird

scripted = torch.jit.script(m)

dhruvbird

m = AddtensorsModel()
res = m(torch.Tensor([1, 2, 3]), torch.Tensor([3, 4, 5]))
print(res)

dhruvbird

import torch

class AddTensorsModel(torch.nn.Module):
  def __init__(self):
    super().__init__();
    self.t1 = torch.Tensor([0, 5, 10])

  def helper(self, x, y):
    z = self.t1
    z = z + x + y