masahi commented on a change in pull request #4497: [WIP] [Relay] Add a PyTorch to Relay Parser URL: https://github.com/apache/incubator-tvm/pull/4497#discussion_r369560699
########## File path: python/tvm/relay/frontend/pytorch.py ########## @@ -0,0 +1,1138 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +# pylint: disable=import-self, too-many-lines, len-as-condition, no-else-return, unused-variable, too-many-nested-blocks +# pylint: disable=consider-iterating-dictionary, invalid-name, unused-argument, unused-variable, broad-except +"""PT: PyTorch frontend.""" +import numpy as np + +import tvm + +from .. import analysis as _analysis +from .. import expr as _expr +from .. import module as _module +from .. import op as _op +from .common import get_relay_op +from .common import infer_shape as _infer_shape + +__all__ = ['from_pytorch'] + +# operator implementation +def _elemwise(name): + def _impl(inputs, input_types): + data0 = convert_input(inputs[0]) + data1 = convert_input(inputs[1]) + + if not isinstance(data0, (_expr.Call, _expr.TupleGetItem, _expr.Var)): + temp = data0 + data0 = data1 + data1 = temp + + return get_relay_op(name)(data0, data1) + return _impl + +def _unsqueeze(): + def _impl(inputs, input_types): + data = inputs[0] + axis = inputs[1] + + return _op.transform.expand_dims(data, int(axis), 1) + return _impl + +def _concatenate(): + def _impl(inputs, input_types): + data = inputs[0] + axis = inputs[1] + + if isinstance(data, (_expr.Call, _expr.TupleGetItem, _expr.Var)): + data = [data] + + return _op.tensor.concatenate(data, int(axis)) + return _impl + +def _slice(): + def _impl(inputs, input_types): + data = inputs[0] + strides = [] + + inferred_shape = _infer_shape(data) + end = [] + for infer in inferred_shape: + end.append(int(infer)) + if isinstance(data, _expr.Var): + end = _infer_shape(data) + end = list(end) + + begin = [0]*len(end) + dim = int(inputs[1]) + begin[dim] = int(inputs[2]) + + if inputs[3].isdigit(): + end[dim] = min(end[dim], int(inputs[3])) + + strides.append(int(inputs[4])) + return _op.transform.strided_slice(data, begin, end, strides) + return _impl + +def _select(): + def _impl(inputs, input_types): + data = inputs[0] + inferred_shape = _infer_shape(data) + end = [] + + for infer in inferred_shape: + end.append(int(infer)) + + begin = [0]*len(end) + dim = int(inputs[1]) + index = int(inputs[2]) + + end[dim] = index+1 + begin[dim] = index + + strides = [1]*len(end) + + sym = _op.transform.strided_slice(data, begin, end, strides) + axis = [dim] + + return _op.transform.squeeze(sym, axis) + return _impl + +def _convert_data_type(input_type): + if input_type == 'double' or input_type == 'torch.float64': + return 'float64' + elif input_type == 'float' or input_type == 'torch.float32': + return 'float32' + elif input_type == 'half' or input_type == 'torch.float16': + return 'float16' + elif input_type == 'long' or input_type == 'torch.int64': + return 'int64' + elif input_type == 'int' or input_type == 'torch.int32': + return 'int32' + elif input_type == 'short' or input_type == 'torch.int16': + return 'int16' + elif input_type == 'char' or input_type == 'torch.int8': + return 'int8' + elif input_type == 'byte' or input_type == 'torch.uint8': + return 'uint8' + else: + return input_type + +def _ones(): + def _impl(inputs, input_types): + if isinstance(inputs[0], _expr.Var): + shape = _infer_shape(inputs[0]) + elif isinstance(inputs[0], (_expr.Call, _expr.TupleGetItem)): + shape = _infer_shape(inputs[0]) + else: + shape = inputs[0].shape + + fill_value = _get_fill_value(input_types) + + return get_relay_op('full')(fill_value, shape, dtype=_convert_data_type(input_types[0])) + return _impl + +def _zeros(): + def _impl(inputs, input_types): + if isinstance(inputs[0], _expr.Var): + shape = _infer_shape(inputs[0]) + elif isinstance(inputs[0], (_expr.Call, _expr.TupleGetItem)): + shape = _infer_shape(inputs[0]) + else: + shape = inputs[0].shape + + fill_value = _get_fill_value(input_types) + + return _op.full(fill_value, shape, dtype=input_types[0]) + return _impl + +def _get_fill_value(input_types): + if input_types[0] == 'int': + fill_value = _expr.const(1) + elif input_types[0] == 'float': + fill_value = _expr.const(1.0) + else: + fill_value = _expr.const(1.0) + + return fill_value + +def _relu(): + def _impl(inputs, input_types): + data = inputs[0] + return _op.nn.relu(data) + return _impl + +def _adaptive_avg_2d(): + def _impl(inputs, input_types): + data = inputs[0] + output_size = _infer_shape(inputs[1]) + + return _op.contrib.contrib.adaptive_avg_pool2d( + data, + output_size=output_size) + return _impl + +def _adaptive_max_2d(): + def _impl(inputs, input_types): + data = inputs[0] + output_size = _infer_shape(inputs[1]) + + return _op.contrib.contrib.adaptive_max_pool2d( + data, + output_size=output_size) + return _impl + +def _maxpool_2d(): + def _impl(inputs, input_types): + data = inputs[0] + + pool_size = _infer_shape(inputs[1]) + strides = _infer_shape(inputs[2]) + padding = _infer_shape(inputs[3]) + + ceil_mode = int(inputs[5]) + + return _op.nn.max_pool2d(data, pool_size, strides, padding, "NCHW", ceil_mode) + return _impl + +def _hardtanh(): + def _impl(inputs, input_types): + a = inputs[0] + tanh_min = float(inputs[1]) + tanh_max = float(inputs[2]) + return _op.tensor.clip(a, tanh_min, tanh_max) + return _impl + +def _convolution(): + def _impl(inputs, input_types): + # Use transpose or normal + use_transpose = False + if inputs[6] == '1': + use_transpose = True + + use_bias = False + if isinstance(inputs[2], _expr.Var): + use_bias = True + + data = inputs[0] + weight = inputs[1] + bias = inputs[2] + + if isinstance(weight, (_expr.Call, _expr.Var, _expr.TupleGetItem)): + inferred_shape = _infer_shape(weight) + weight_shape = [] + for infer in inferred_shape: + weight_shape.append(infer) + else: + weight_shape = weight.shape + channels = weight_shape[0] + + strides = inputs[3] + padding = inputs[4] + dilation = inputs[5] + + kernel_size = weight_shape[2:] + + else: + data = inputs[0] + weight = inputs[1] + bias = inputs[2] + + if isinstance(weight, (_expr.Call, _expr.Var, _expr.TupleGetItem)): + inferred_shape = _infer_shape(weight) + weight_shape = [] + for infer in inferred_shape: + weight_shape.append(infer) + else: + weight_shape = weight.shape + channels = weight_shape[0] + + strides = inputs[3] + padding = inputs[4] + dilation = inputs[5] + + kernel_size = weight_shape[2:] + + if isinstance(strides, _expr.Var): + strides = _infer_shape(strides) + + if isinstance(padding, _expr.Var): + padding = _infer_shape(padding) + + if isinstance(dilation, _expr.Var): + dilation = _infer_shape(dilation) + + groups = int(inputs[8]) + + if use_transpose: + conv_out = _op.nn.conv2d_transpose(data, + weight, + strides=strides, + padding=padding, + dilation=dilation, + groups=groups, + channels=channels, + kernel_size=kernel_size, + data_layout="NCHW", + kernel_layout="OIHW", + out_layout="", + out_dtype="") + else: + conv_out = _op.nn.conv2d(data, + weight, + strides=strides, + padding=padding, + dilation=dilation, + groups=groups, + channels=channels, + kernel_size=kernel_size, + data_layout="NCHW", + kernel_layout="OIHW", + out_layout="", + out_dtype="") + + if use_bias: + return _op.nn.bias_add(conv_out, bias) + else: + return conv_out + return _impl + +def _softmax(): + def _impl(inputs, input_types): + data = inputs[0] + axis = inputs[1] + if isinstance(axis, str): + axis = int(axis) + + return _op.nn.softmax(data, axis=axis) + return _impl + +def _threshold(): + def _impl(inputs, input_types): + data = inputs[0] + return _op.nn.relu(data) + return _impl + +def _contiguous(): + def _impl(inputs, input_types): + data = inputs[0] + return _op.tensor.copy(data) + return _impl + +def _batch_norm(): + def _impl(inputs, input_types): + data = inputs[0] + data_type = input_types[0] + + channels = _infer_shape(data) + + if isinstance(inputs[1], _expr.Var) and isinstance(inputs[2], _expr.Var): + scale = center = True + weight = inputs[1] + beta = inputs[2] + else: + scale = center = False + + if scale: + gamma = weight + else: + if data_type == 'double': + gamma = _expr.const(np.ones([int(channels[1])]).astype('float64')) + elif data_type == 'float': + gamma = _expr.const(np.ones([int(channels[1])]).astype('float32')) + elif data_type == 'half': + gamma = _expr.const(np.ones([int(channels[1])]).astype('float16')) + elif data_type == 'long': + gamma = _expr.const(np.ones([int(channels[1])]).astype('int64')) + elif data_type == 'int': + gamma = _expr.const(np.ones([int(channels[1])]).astype('int32')) + elif data_type == 'short': + gamma = _expr.const(np.ones([int(channels[1])]).astype('int16')) + elif data_type == 'char': + gamma = _expr.const(np.ones([int(channels[1])]).astype('int8')) + elif data_type == 'byte': + gamma = _expr.const(np.ones([int(channels[1])]).astype('uint8')) + + if center: + beta = beta + else: + if data_type == 'double': + beta = _expr.const(np.zeros([int(channels[1])]).astype('float64')) + elif data_type == 'float': + beta = _expr.const(np.zeros([int(channels[1])]).astype('float32')) + elif data_type == 'half': + beta = _expr.const(np.zeros([int(channels[1])]).astype('float16')) + elif data_type == 'long': + beta = _expr.const(np.zeros([int(channels[1])]).astype('int64')) + elif data_type == 'int': + beta = _expr.const(np.zeros([int(channels[1])]).astype('int32')) + elif data_type == 'short': + beta = _expr.const(np.zeros([int(channels[1])]).astype('int16')) + elif data_type == 'char': + beta = _expr.const(np.zeros([int(channels[1])]).astype('int8')) + elif data_type == 'byte': + beta = _expr.const(np.zeros([int(channels[1])]).astype('uint8')) + + moving_mean = inputs[3] + moving_var = inputs[4] + epsilon = float(inputs[7]) + + center = center + scale = scale + + return _op.nn.batch_norm(data, + gamma, + beta, + moving_mean, + moving_var, + axis=1, + epsilon=epsilon, + center=center, + scale=scale)[0] + return _impl + +def _transpose(): + def _impl(inputs, input_types): + data = inputs[0] + + if isinstance(data, _expr.Var): + ndims = len(_infer_shape(data)) + elif isinstance(data, (_expr.Call, _expr.TupleGetItem)): + ndims = _infer_shape(data) + else: + ndims = data.shape + + if isinstance(data, tvm.ndarray.NDArray): + ndims = len(data.shape) + axes = list(range(ndims)) + + num_inputs = len(inputs) + + if num_inputs == 1: + if ndims >= 2: + axes[-1] = ndims - 2 + axes[-2] = ndims - 1 + if not isinstance(data, _expr.Var): + data = _expr.const(data) + + elif num_inputs == 3: + parse = lambda i: ndims * (i < 0) + i + src, dst = [parse(int(inputs[i])) for i in [1, 2]] + axes[src] = dst + axes[dst] = src + else: + axes = inputs[1] + return _op.transform.transpose(data, axes) + return _impl + +def _flatten(): + def _impl(inputs, input_types): + data = inputs[0] + return _op.nn.batch_flatten(data) + return _impl + +def _dense(): + def _impl(inputs, input_types): + use_bias = False + + if isinstance(inputs[0], _expr.Var): + use_bias = True + + data = inputs[1] + data_type = input_types[1] + weight = inputs[2] + + beta = inputs[3] + alpha = inputs[4] + + if not isinstance(alpha, (_expr.Var, _expr.Call, _expr.TupleGetItem)): + if data_type == 'double': + alpha = _expr.const(np.float64(alpha), dtype='float64') + elif data_type == 'float': + alpha = _expr.const(np.float32(alpha), dtype='float32') + elif data_type == 'half': + alpha = _expr.const(np.float16(alpha), dtype='float16') + elif data_type == 'long': + alpha = _expr.const(np.int64(alpha), dtype='int64') + elif data_type == 'int': + alpha = _expr.const(np.int32(alpha), dtype='int32') + elif data_type == 'short': + alpha = _expr.const(np.int16(alpha), dtype='int16') + elif data_type == 'char': + alpha = _expr.const(np.int8(alpha), dtype='int8') + elif data_type == 'byte': + alpha = _expr.const(np.uint8(alpha), dtype='uint8') + data *= alpha + + if not isinstance(beta, (_expr.Var, _expr.Call, _expr.TupleGetItem)): + if data_type == 'double': + beta = _expr.const(np.float64(beta), dtype='float64') + elif data_type == 'float': + beta = _expr.const(np.float32(beta), dtype='float32') + elif data_type == 'half': + beta = _expr.const(np.float16(beta), dtype='float16') + elif data_type == 'long': + beta = _expr.const(np.int64(beta), dtype='int64') + elif data_type == 'int': + beta = _expr.const(np.int32(beta), dtype='int32') + elif data_type == 'short': + beta = _expr.const(np.int16(beta), dtype='int16') + elif data_type == 'char': + beta = _expr.const(np.int8(beta), dtype='int8') + elif data_type == 'byte': + beta = _expr.const(np.uint8(beta), dtype='uint8') + weight *= beta + + weight_out = _op.transform.transpose(weight, axes=[1, 0]) + + units = _infer_shape(weight_out)[0] + dense_out = _op.nn.dense(data, weight_out, units=units) + + if use_bias: + bias = inputs[0] + return _op.nn.bias_add(dense_out, bias) + else: + return dense_out + return _impl + +def _size(): + def _impl(inputs, input_types): + axis = int(inputs[1]) + if isinstance(inputs[0], _expr.Var): + shape = _infer_shape(inputs[0]) + else: + shape = _infer_shape(inputs[0]) + return shape[axis] + return _impl + +def _numtotensor(): + def _impl(inputs, input_types): + val = inputs[0] + dtype = type(val) + + if isinstance(val, tvm.expr.IntImm): + val = val.__int__() + dtype = int + + arr = val * np.ones([]).astype(dtype) + return arr + return _impl + +def _view(): + def _impl(inputs, input_types): + data = inputs[0] + + if len(inputs) == 3: + new_shape = [inputs[1], _infer_shape(inputs[2])[0]] + else: + if isinstance(inputs[1], list): + new_shape = inputs[1] + else: + new_shape = _infer_shape(inputs[1]) + + return _op.transform.reshape(data, new_shape) + return _impl + +def _clone(): + def _impl(inputs, input_types): + data = inputs[0] + return _op.tensor.copy(data) + return _impl + +def _log_softmax(): + def _impl(inputs, input_types): + data = inputs[0] + axis = int(inputs[1]) + return _op.nn.log_softmax(data, axis) + return _impl + +def _sigmoid(): + def _impl(inputs, input_types): + data = inputs[0] + return _op.tensor.sigmoid(data) + return _impl + +def _avg_pool2d(): + def _impl(inputs, input_types): + data = inputs[0] + + pool_size = _infer_shape(inputs[1]) + strides = _infer_shape(inputs[2]) + padding = _infer_shape(inputs[3]) + + ceil_mode = int(inputs[4]) + count_include_pad = int(inputs[5]) + + return _op.nn.avg_pool2d(data, + pool_size=pool_size, + strides=strides, + padding=padding, + ceil_mode=ceil_mode, + count_include_pad=count_include_pad) + return _impl + +def _dropout(): + def _impl(inputs, input_types): + data = inputs[0] + rate = float(inputs[1]) + + return _op.nn.dropout(data, rate) + return _impl + +def _reduce(name): + def _impl(inputs, attrs, params): + data = inputs[0] + return get_relay_op(name)(data) + return _impl + +def _mean(): + def _impl(inputs, input_types): + data = inputs[0] + axis = _infer_shape(inputs[1]) + + keepdims = int(inputs[2]) + exclude = int(inputs[3]) + + return _op.mean(data, axis, keepdims, exclude) + return _impl + +def _chunk(): + def _impl(inputs, input_types): + data = inputs[0] + + num_chunks = int(inputs[1]) + axis = int(inputs[2]) + + if isinstance(data, _expr.Var): + inferred_shape = _infer_shape(data) + elif isinstance(data, (_expr.Call, _expr.TupleGetItem)): + inferred_shape = _infer_shape(data) + + shape = [] + for infer in inferred_shape: + shape.append(infer) + + dim = int(shape[axis]) + + if dim % num_chunks: + unif_size = int(dim / (num_chunks - 1)) + else: + unif_size = int(dim / num_chunks) + + chunks = [] + for i in range(0, dim, unif_size): + begin = [0] * len(shape) + end = shape[:] + begin[axis] = i + end[axis] = i + unif_size + stride = [1] * len(shape) + + chunk_out = _op.transform.strided_slice(data, begin, end, stride) + chunks.append(chunk_out) + + + if dim % num_chunks: + begin = [0] * len(shape) + end = shape[:] + begin[axis] = unif_size * (num_chunks - 1) + end[axis] = dim + stride = [1] * len(shape) + + chunk_out = _op.transform.strided_slice(data, begin, end, stride) + chunks.append(chunk_out) + + return chunks + return _impl + +def _matmul(): + def _impl(inputs, input_types): + data0 = inputs[0] + data1 = inputs[1] + data1_t = _op.transpose(data1, axes=(1, 0)) + + return _op.nn.dense(data0, data1_t) + return _impl + +def _expand(): + def _impl(inputs, input_types): + data_in = inputs[0] + if isinstance(data_in, _expr.Var): + shape = _infer_shape(data_in) + elif isinstance(data_in, (_expr.Call, _expr.TupleGetItem)): + shape = _infer_shape(data_in) + + ndims = len(shape) + sizes = _infer_shape(inputs[1]) + out = inputs[0] + + for i in range(ndims): + if sizes[i] in {-1, shape[i]}: + continue + data = list() + for temp in range(sizes[i]): + data.append(out) + call = _op.tensor.concatenate(data, i) + + return call + return _impl + +def _int(): + def _impl(inputs, input_types): + if isinstance(inputs[0], _expr.Call): + return inputs[0] + return int(inputs[0]) + return _impl + +def _listunpack(): + def _impl(inputs, input_types): + return inputs[0] + return _impl + +def _to(): + def _impl(inputs, input_types): + return inputs[0] + return _impl + +def _device(): + def _impl(inputs, input_types): + return None + return _impl + +def _pad(): + def _impl(inputs, input_types): + data = inputs[0] + padding = inputs[1] + pad_width = list(zip(padding, padding)) + pad_value = inputs[2] + return _op.nn.pad(data, pad_width, pad_value) + return _impl + +def _sqrt(): + def _impl(inputs, input_types): + data = inputs[0] + return _op.tensor.sqrt(data) + return _impl + +# Helper functions for operator implementation + +def convert_input(data): + """ Handle input conversion for elemwise op """ + if isinstance(data, (_expr.Call, _expr.TupleGetItem, _expr.Var)): + return data + elif isinstance(data, str): + if len(data) == 1: + return _expr.const(int(data), dtype='float32') + else: + if '.' in data: + return _expr.const(float(data[1:-1]), dtype='float32') + else: + return _expr.const(int(data[1:-1]), dtype='float32') + else: + return _expr.const(int(data), dtype='float32') + +# Operator mappings + +_convert_map = { + 'aten::device' : _device(), + 'aten::add' : _elemwise('add'), + 'aten::add_' : _elemwise('add'), + 'aten::sub' : _elemwise('subtract'), + 'aten::sub_' : _elemwise('subtract'), + 'aten::max' : _elemwise('maximum'), + 'aten::min' : _elemwise('minimum'), + 'aten::mul' : _elemwise('multiply'), + 'aten::mul_' : _elemwise('multiply'), + 'aten::pow' : _elemwise('power'), + 'aten::div' : _elemwise('divide'), + 'aten::div_' : _elemwise('divide'), + 'aten::ones' : _ones(), + 'aten::zeros' : _zeros(), + 'aten::to' : _to(), + 'aten::unsqueeze' : _unsqueeze(), + 'aten::cat' : _concatenate(), + 'aten::slice' : _slice(), + 'aten::select' : _select(), + 'aten::relu' : _relu(), + 'aten::relu_' : _relu(), + 'aten::adaptive_avg_pool2d' : _adaptive_avg_2d(), + 'aten::adaptive_max_pool2d' : _adaptive_max_2d(), + 'aten::max_pool2d' : _maxpool_2d(), + 'aten::max_pool2d_with_indices' : _maxpool_2d(), + 'aten::hardtanh' : _hardtanh(), + 'aten::hardtanh_' : _hardtanh(), + 'aten::_convolution' : _convolution(), + 'aten::softmax' : _softmax(), + 'aten::threshold' : _threshold(), + 'aten::threshold_' : _threshold(), + 'aten::contiguous' : _contiguous(), + 'aten::batch_norm' : _batch_norm(), + 'aten::transpose' : _transpose(), + 'aten::transpose_' : _transpose(), + 'aten::t' : _transpose(), + 'aten::flatten' : _flatten(), + 'aten::addmm' : _dense(), + 'aten::size' : _size(), + 'aten::view' : _view(), + 'aten::clone' : _clone(), + 'aten::log_softmax' : _log_softmax(), + 'aten::sigmoid' : _sigmoid(), + 'aten::avg_pool2d' : _avg_pool2d(), + 'aten::dropout' : _dropout(), + 'aten::dropout_' : _dropout(), + 'aten::mean' : _mean(), + 'aten::chunk' : _chunk(), + 'aten::matmul' : _matmul(), + 'aten::expand' : _expand(), + 'aten::Int' : _int(), + 'prim::NumToTensor' : _numtotensor(), + 'prim::ListUnpack' : _listunpack(), + 'aten::constant_pad_nd' : _pad(), + 'aten::permute' : _transpose(), + 'aten::sum' : _reduce('sum'), + 'aten::prod' : _reduce('prod'), + 'aten::sqrt' : _sqrt() +} + +# Internal graph for parsing + +class Graph(object): + """ A helper class for handling relay graph copying from PyTorch trace. """ + + def __init__(self, trace, input_shapes, input_types): + + self._trace = trace + self._inputs_r = {} + self._params = {} + self._param_tensors = {} + self._consts = {} + self._ops = {} + self._op_inputs_r = {} + self._op_inputs_types = {} + self._op_inputs_otypes = {} + self._input_shapes = input_shapes if input_shapes else {} + self._input_types = input_types if input_types else {} + self._fn_param = [] + self._relay_map = {} + self._nid_to_node_name = {} + + def from_pytorch(self): + """ Construct relay nodes from trace of PyTorch graph + + Currently only supports traced PyTorch format which means no control flow. + User must perform torch.jit.trace on a model and pass this in. + Future support should include support scripted models (torch.jit.script) which + preserves control flow. + + Returns + ------- + mod : tvm.relay.Module + The module that optimizations will be performed on. + + params : dict of str to tvm.ndarray + Dict of converted parameters stored in tvm.ndarray format + """ + # Check for missing ops + missing_operators = self._parse_import_prerequisites() + + if missing_operators: + raise NotImplementedError( \ + "The following operators are not implemented: {}".format(missing_operators)) + + # Translate PyTorch graph to by decorating Graph with state dict and inputs into each op + self._parse_inputs() + self._parse_params() + self._parse_ops() + + nid = 0 + for (op_name, operator), op_node in self._ops.items(): + if operator == 'prim::Constant': + pass + elif operator == 'prim::ListConstruct': + if any(inp.debugName() in self._nid_to_node_name.keys() \ + for inp in op_node.inputs()): + listconstr = [] + for i in op_node.inputs(): + if i.debugName() in self._nid_to_node_name.keys(): + listconstr.append( \ + self._relay_map[self._nid_to_node_name[i.debugName()]]) + elif i.node().kind() == 'prim::Constant': + listconstr.append(int(self._consts[i.debugName()])) + elif i.debugName() in self._inputs_r.keys(): + listconstr.append(int(self._inputs_r[i.debugName()])) + + # Unwrap for tensors + if len(listconstr) == 1: + listconstr = listconstr[0] + + self._relay_map[nid] = listconstr + self._nid_to_node_name[op_name] = nid + nid = nid+1 + else: + for i in op_node.inputs(): + if i.debugName() in self._nid_to_node_name.keys(): + for cnt in range(0, len(self._op_inputs_r[(op_name, operator)])): + if isinstance(self._op_inputs_r[(op_name, operator)][cnt], str): + if "call/var" in self._op_inputs_r[(op_name, operator)][cnt]: + self._op_inputs_r[(op_name, operator)][cnt] = \ + self._relay_map[self._nid_to_node_name[i.debugName()]] + break + + call = _convert_map[operator](self._op_inputs_r[(op_name, operator)], + self._op_inputs_types[(op_name, operator)]) + + self._relay_map[nid] = call + self._nid_to_node_name[op_name] = nid + nid = nid+1 + + outputs = [] + + for i in range(nid): + output = self._relay_map[i] + outputs.append(output) + + if len(outputs) == 1: + body = outputs[0] + else: + body = outputs[-1] + + func = tvm.relay.Function(_analysis.free_vars(body), body) + + param = {k: tvm.nd.array(v) for k, v in self._param_tensors.items()} + + return _module.Module.from_expr(func), param + + def _parse_inputs(self): + """ Map inputs to parser and inputs to graph. """ + # Get names and objects of inputs for IR + ir_names = [i.debugName() for i in self._trace.graph.inputs()] + ir_inputs = [i for i in self._trace.graph.inputs()] + + # Create corresponding shape and add to input + for input_name, ir_input in zip(self._input_shapes, ir_inputs[1:]): + input_shape = self._input_shapes[input_name] + ir_input.setDebugName(input_name) + + self._inputs_r[input_name] = _expr.var(input_name, + shape=self._input_shapes[input_name], + dtype=_convert_data_type( + self._input_types[input_name])) + self._fn_param.append(_expr.var(input_name, + shape=self._input_shapes[input_name], + dtype=_convert_data_type( + self._input_types[input_name]))) + + # Add self (first input of a PyTorch graph) to inputs + input_shape = [3] + tensor = tvm.nd.array(np.zeros(input_shape).astype(np.float32)) + input_name = ir_names[0] + self._inputs_r[input_name] = tensor + + def _parse_params(self): + """ Map state dictionary values to corresponding prim::GetAttr op node. """ + # Grab weights, biases, etc. from graph + state_dict = self._trace.state_dict() + param_names = [] + for key, value in state_dict.items(): + param_str = str(key) + param_name = param_str.split('.')[-1] + param_names.append(param_name) + + # Get names of all inputs + input_names = [i for i in self._inputs_r.keys()] + + # Iterate through graph for getAttr nodes and match full state_dict name to nodes + node_weight_map = {} + for node in self._trace.graph.nodes(): + if node.kind() == "prim::GetAttr": + node_str = str(node) + node_assign = (node_str.split(' = ')[0]).split(' : ') + node_name = (node_assign[0])[1:] + node_getattr_name = ((node_str.split(' = ')[1]).split('"')[1::2])[0] + node_arg = (((node_str.split(' = '))[1]).split('(')[1])[1:-2] + + if node_arg in input_names: + node_weight_map[node_name] = node_getattr_name + else: + previous_map = node_weight_map[node_arg[:]] + node_weight_map[node_name] = previous_map+"."+node_getattr_name + + if node_getattr_name in param_names: + + value = state_dict[node_weight_map[node_name]] + tensor = tvm.nd.array(value.cpu().numpy()) + shape = tensor.shape + self._param_tensors[node_name] = tensor + + self._params[node_name] = _expr.var(node_name, + shape=shape, + dtype=_convert_data_type(str(value.dtype))) + + self._fn_param.append(_expr.var(node_name, + shape=shape, + dtype=_convert_data_type(str(value.dtype)))) + + + def _parse_ops(self): + """ Iterate through nodes and decorate graph with constants, operators, + and the inputs to each operator. """ + # Traverse nodes and add to graph + for node in self._trace.graph.nodes(): + + node_str = str(node) + node_assign = (node_str.split(' = ')[0]).split(' : ') + node_name = (node_assign[0])[1:] + node_expr = (node_str.split(' = ')[1]).split(',')[0] + + if node.kind() == "prim::Constant": Review comment: ``` if node.kind() == "prim::Constant" and len(node.attributeNames()) == 1: attribute_names = node.attributeNames() attr_name = attribute_names[0] ty = node.output().type().kind() if ty == "IntType" or ty == "BoolType": consts[node_name] = node.i(attr_name) elif ty == "FloatType": consts[node_name] = node.f(attr_name) ``` you may need to handle other types ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services