tvm onnx import error, using PyTorch models.

tvm_onnx_model_error.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch \n",
    "import torch.nn as nn\n",
    "import torchvision\n",
    "import torchvision.transforms as transforms\n",
    "from torch.autograd import Variable\n",
    "import numpy as np\n",
    "import torch.onnx\n",
    "\n",
    "import tvm\n",
    "import tvm.relay as relay\n",
    "from matplotlib import pyplot as plt\n",
    "from tvm.contrib import graph_runtime\n",
    "import onnx"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Generate PyTorch model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ConvNet(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(ConvNet, self).__init__()\n",
    "        self.layer1 = nn.Sequential(\n",
    "            nn.Conv2d(in_c, num_filters, kernel_size=kdim, \n",
    "                      stride=stride, padding=padding, bias=False))    \n",
    "    def forward(self, x):\n",
    "        out = self.layer1(x)\n",
    "        return out"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set hyperparams\n",
    "in_c = 512\n",
    "num_filters = 512\n",
    "in_h, in_w = 28, 28\n",
    "kdim = 3\n",
    "groups=1\n",
    "stride=1\n",
    "padding=0\n",
    "in_shape = (1, in_c, in_h, in_w)\n",
    "\n",
    "test_input = Variable(torch.randn(1, in_c, in_h, in_w))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create model, and export to onnx\n",
    "save_path = '/tmp/'\n",
    "fname = 'mymodel'\n",
    "\n",
    "save_name = save_path + fname + '.onnx'\n",
    "pytorch_model = ConvNet()\n",
    "torch.onnx.export(pytorch_model, test_input, save_name)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load Model into tvm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "graph torch-jit-export (\n",
      "  %0[FLOAT, 1x512x28x28]\n",
      ") initializers (\n",
      "  %1[FLOAT, 512x512x3x3]\n",
      ") {\n",
      "  %2 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%0, %1)\n",
      "  return %2\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "onnx_model = onnx.load(save_name)\n",
    "# Check that the IR is well formed\n",
    "onnx.checker.check_model(onnx_model)\n",
    "# Print a human readable representation of the graph\n",
    "print(onnx.helper.printable_graph(onnx_model.graph))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create relay dummy input\n",
    "test_input = np.random.rand(1, in_c, in_h, in_w)\n",
    "data   = relay.var(\"data\", relay.TensorType(in_shape, \n",
    "                                            \"float32\"))\n",
    "dtype = 'float32'\n",
    "\n",
    "x = test_input\n",
    "ctx = tvm.cpu()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# compile with relay\n",
    "\n",
    "input_name = '1'\n",
    "shape_dict = {input_name: x.shape}\n",
    "target = 'llvm'\n",
    "\n",
    "sym, params = relay.frontend.from_onnx(onnx_model, shape_dict)\n",
    "\n",
    "with relay.build_config(opt_level=1):\n",
    "    intrps = relay.build_module.create_executor('graph', sym, \n",
    "                                                tvm.cpu(0), target)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "ename": "TVMError",
     "evalue": "Traceback (most recent call last):\n  [bt] (8) /home/wheest/tools/tvm/build/libtvm.so(TVMFuncCall+0x4a) [0x7f3eb62dd95a]\n  [bt] (7) /home/wheest/tools/tvm/build/libtvm.so(+0x71b094) [0x7f3eb60b1094]\n  [bt] (6) /home/wheest/tools/tvm/build/libtvm.so(tvm::relay::InferType(tvm::relay::Expr const&, tvm::relay::Module const&)+0x2b4) [0x7f3eb60b0634]\n  [bt] (5) /home/wheest/tools/tvm/build/libtvm.so(+0x4bd479) [0x7f3eb5e53479]\n  [bt] (4) /home/wheest/tools/tvm/build/libtvm.so(+0x4bbf53) [0x7f3eb5e51f53]\n  [bt] (3) /home/wheest/tools/tvm/build/libtvm.so(tvm::relay::InferType(tvm::relay::Function const&, tvm::relay::Module const&, tvm::relay::GlobalVar const&)+0x1a2) [0x7f3eb60b0b32]\n  [bt] (2) /home/wheest/tools/tvm/build/libtvm.so(+0x71a003) [0x7f3eb60b0003]\n  [bt] (1) /home/wheest/tools/tvm/build/libtvm.so(+0x48ca1f) [0x7f3eb5e22a1f]\n  [bt] (0) /home/wheest/tools/tvm/build/libtvm.so(+0x10d852) [0x7f3eb5aa3852]\n  [bt] (8) /home/wheest/tools/tvm/build/libtvm.so(+0x719fe7) [0x7f3eb60affe7]\n  [bt] (7) /home/wheest/tools/tvm/build/libtvm.so(+0x73373b) [0x7f3eb60c973b]\n  [bt] (6) /home/wheest/tools/tvm/build/libtvm.so(+0x736d5d) [0x7f3eb60ccd5d]\n  [bt] (5) /home/wheest/tools/tvm/build/libtvm.so(+0x5089ad) [0x7f3eb5e9e9ad]\n  [bt] (4) /home/wheest/tools/tvm/build/libtvm.so(+0x508a51) [0x7f3eb5e9ea51]\n  [bt] (3) /home/wheest/tools/tvm/build/libtvm.so(+0x5249a2) [0x7f3eb5eba9a2]\n  [bt] (2) /home/wheest/tools/tvm/build/libtvm.so(tvm::BijectiveLayout::ForwardShape(tvm::Array<HalideIR::Expr, void> const&) const+0xbd) [0x7f3eb5c0359d]\n  [bt] (1) /home/wheest/tools/tvm/build/libtvm.so(+0x26e7f7) [0x7f3eb5c047f7]\n  [bt] (0) /home/wheest/tools/tvm/build/libtvm.so(+0x10d852) [0x7f3eb5aa3852]\n  File \"/home/wheest/tools/tvm/src/relay/ir/error.cc\", line 131\nTVMError: \u001b[1m\nError(s) have occurred. We have annotated the program with them:\n\n\u001b[0m\u001b[1mIn `main`: \n\u001b[0mv0.0.1\n%1 = fn (%v0: float32, %v1: Tensor[(512, 512, 3, 3), float32]) {\n  %0 = nn.conv2d(%v0, %v1, kernel_size=[3, 3]) // \u001b[31man internal invariant was violdated while typechecking your program [12:20:33] /home/wheest/tools/tvm/src/lang/data_layout.cc:273: Check failed: src_shape.size() == src_axis.size() (0 vs. 4) : \n; \u001b[39m\n  %0\n}\n%1\n",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mTVMError\u001b[0m                                  Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-8-258695ce744b>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;31m# execute in tvm\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0mtvm_output\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mintrps\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mevaluate\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0msym\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minput_data\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mparams\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0masnumpy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m~/tools/tvm/python/tvm/relay/backend/interpreter.py\u001b[0m in \u001b[0;36mevaluate\u001b[0;34m(self, expr, binds)\u001b[0m\n\u001b[1;32m    225\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    226\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mexpr\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mFunction\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mGlobalVar\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 227\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_make_executor\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mexpr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    228\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    229\u001b[0m         \u001b[0;31m# normal expression evaluated by running a function.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/tools/tvm/python/tvm/relay/build_module.py\u001b[0m in \u001b[0;36m_make_executor\u001b[0;34m(self, func)\u001b[0m\n\u001b[1;32m    428\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    429\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m_make_executor\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 430\u001b[0;31m         \u001b[0mgraph_json\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmod\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mparams\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mbuild\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfunc\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtarget\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtarget\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    431\u001b[0m         \u001b[0mgmodule\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_graph_rt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcreate\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mgraph_json\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmod\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mctx\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    432\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mparams\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/tools/tvm/python/tvm/relay/build_module.py\u001b[0m in \u001b[0;36mbuild\u001b[0;34m(func, target, target_host, params)\u001b[0m\n\u001b[1;32m    290\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    291\u001b[0m     \u001b[0;32mwith\u001b[0m \u001b[0mtophub_context\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 292\u001b[0;31m         \u001b[0mfunc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0moptimize\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfunc\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtarget\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mparams\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    293\u001b[0m         \u001b[0;31m# Annotate the ops for heterogeneous execution.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    294\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtarget\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdict\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/tools/tvm/python/tvm/relay/build_module.py\u001b[0m in \u001b[0;36moptimize\u001b[0;34m(func, target, params)\u001b[0m\n\u001b[1;32m    177\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    178\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mcfg\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpass_enabled\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"SimplifyInference\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 179\u001b[0;31m         \u001b[0mfunc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mir_pass\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0minfer_type\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfunc\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    180\u001b[0m         \u001b[0mfunc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mir_pass\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msimplify_inference\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfunc\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    181\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/tools/tvm/python/tvm/relay/ir_pass.py\u001b[0m in \u001b[0;36minfer_type\u001b[0;34m(expr, mod)\u001b[0m\n\u001b[1;32m    367\u001b[0m         \u001b[0mThe\u001b[0m \u001b[0mchecked\u001b[0m \u001b[0mexpression\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    368\u001b[0m     \"\"\"\n\u001b[0;32m--> 369\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0m_ir_pass\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0minfer_type\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mexpr\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmod\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    370\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    371\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/tools/tvm/python/tvm/_ffi/_ctypes/function.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *args)\u001b[0m\n\u001b[1;32m    204\u001b[0m                 \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mhandle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalues\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtcodes\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mctypes\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mc_int\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnum_args\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    205\u001b[0m                 ctypes.byref(ret_val), ctypes.byref(ret_tcode)) != 0:\n\u001b[0;32m--> 206\u001b[0;31m             \u001b[0;32mraise\u001b[0m \u001b[0mget_last_ffi_error\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    207\u001b[0m         \u001b[0m_\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtemp_args\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    208\u001b[0m         \u001b[0m_\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mTVMError\u001b[0m: Traceback (most recent call last):\n  [bt] (8) /home/wheest/tools/tvm/build/libtvm.so(TVMFuncCall+0x4a) [0x7f3eb62dd95a]\n  [bt] (7) /home/wheest/tools/tvm/build/libtvm.so(+0x71b094) [0x7f3eb60b1094]\n  [bt] (6) /home/wheest/tools/tvm/build/libtvm.so(tvm::relay::InferType(tvm::relay::Expr const&, tvm::relay::Module const&)+0x2b4) [0x7f3eb60b0634]\n  [bt] (5) /home/wheest/tools/tvm/build/libtvm.so(+0x4bd479) [0x7f3eb5e53479]\n  [bt] (4) /home/wheest/tools/tvm/build/libtvm.so(+0x4bbf53) [0x7f3eb5e51f53]\n  [bt] (3) /home/wheest/tools/tvm/build/libtvm.so(tvm::relay::InferType(tvm::relay::Function const&, tvm::relay::Module const&, tvm::relay::GlobalVar const&)+0x1a2) [0x7f3eb60b0b32]\n  [bt] (2) /home/wheest/tools/tvm/build/libtvm.so(+0x71a003) [0x7f3eb60b0003]\n  [bt] (1) /home/wheest/tools/tvm/build/libtvm.so(+0x48ca1f) [0x7f3eb5e22a1f]\n  [bt] (0) /home/wheest/tools/tvm/build/libtvm.so(+0x10d852) [0x7f3eb5aa3852]\n  [bt] (8) /home/wheest/tools/tvm/build/libtvm.so(+0x719fe7) [0x7f3eb60affe7]\n  [bt] (7) /home/wheest/tools/tvm/build/libtvm.so(+0x73373b) [0x7f3eb60c973b]\n  [bt] (6) /home/wheest/tools/tvm/build/libtvm.so(+0x736d5d) [0x7f3eb60ccd5d]\n  [bt] (5) /home/wheest/tools/tvm/build/libtvm.so(+0x5089ad) [0x7f3eb5e9e9ad]\n  [bt] (4) /home/wheest/tools/tvm/build/libtvm.so(+0x508a51) [0x7f3eb5e9ea51]\n  [bt] (3) /home/wheest/tools/tvm/build/libtvm.so(+0x5249a2) [0x7f3eb5eba9a2]\n  [bt] (2) /home/wheest/tools/tvm/build/libtvm.so(tvm::BijectiveLayout::ForwardShape(tvm::Array<HalideIR::Expr, void> const&) const+0xbd) [0x7f3eb5c0359d]\n  [bt] (1) /home/wheest/tools/tvm/build/libtvm.so(+0x26e7f7) [0x7f3eb5c047f7]\n  [bt] (0) /home/wheest/tools/tvm/build/libtvm.so(+0x10d852) [0x7f3eb5aa3852]\n  File \"/home/wheest/tools/tvm/src/relay/ir/error.cc\", line 131\nTVMError: \u001b[1m\nError(s) have occurred. We have annotated the program with them:\n\n\u001b[0m\u001b[1mIn `main`: \n\u001b[0mv0.0.1\n%1 = fn (%v0: float32, %v1: Tensor[(512, 512, 3, 3), float32]) {\n  %0 = nn.conv2d(%v0, %v1, kernel_size=[3, 3]) // \u001b[31man internal invariant was violdated while typechecking your program [12:20:33] /home/wheest/tools/tvm/src/lang/data_layout.cc:273: Check failed: src_shape.size() == src_axis.size() (0 vs. 4) : \n; \u001b[39m\n  %0\n}\n%1\n"
     ]
    }
   ],
   "source": [
    "# execute in tvm\n",
    "tvm_output = intrps.evaluate(sym)(input_data, **params).asnumpy()\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "meth",
   "language": "python",
   "name": "meth"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}