import torch import torch.nn.functional as F import torch.nn as nn import torch.nn.quantized as nnq import torch.nn.quantized.dynamic as nnqd import torch.nn.intrinsic as nni import torch.nn.intrinsic.quantized as nniq import torch.multiprocessing as mp # graph mode quantization based on fx from torch.quantization.quantize_fx import ( prepare_fx, convert_fx, prepare_qat_fx, ) from torch.quantization.fx.pattern_utils import ( is_match, MatchAllNode, ) from torch.quantization import ( QuantType, QuantStub, DeQuantStub, QuantWrapper, quant_type_to_str, default_qconfig, default_dynamic_qconfig, default_qat_qconfig, per_channel_dynamic_qconfig, float16_dynamic_qconfig, float_qparams_weight_only_qconfig, get_default_qconfig, get_default_qat_qconfig, fuse_modules, prepare, prepare_qat, convert, quantize_dynamic, default_placeholder_observer, PerChannelMinMaxObserver, QConfigDynamic, FixedQParamsFakeQuantize, ) # test utils from torch.testing._internal.common_cuda import TEST_MULTIGPU, TEST_CUDA from torch.testing._internal.common_quantization import ( QuantizationTestCase, skipIfNoFBGEMM, skip_if_no_torchvision, train_one_epoch, run_ddp, test_only_eval_fn, test_only_train_fn, ) from torch.testing._internal.common_quantization import ( LinearModelWithSubmodule, ResNetBase, RNNDynamicModel, RNNCellDynamicModel, ) from torch.testing._internal.common_quantized import ( supported_qengines, override_qengines, override_quantized_engine, ) from torch.testing._internal.common_distributed import skip_if_not_multigpu from torch.testing._internal.common_quantization import NodeSpec as ns from torch.testing import FileCheck import copy import itertools import operator import unittest import io from typing import Callable class TestFuseFx(QuantizationTestCase): def test_fuse_conv_bn_relu(self): class M(torch.nn.Module): def __init__(self): super().__init__() self.conv1d = nn.Conv1d(1, 1, 1) self.conv2d = nn.Conv2d(1, 1, 1) self.conv3d = nn.Conv3d(1, 1, 1) self.bn1d = nn.BatchNorm1d(1) self.bn2d = nn.BatchNorm2d(1) self.bn3d = nn.BatchNorm3d(1) self.conv1d2 = nn.Conv1d(1, 1, 1) self.conv2d2 = nn.Conv2d(1, 1, 1) self.conv3d2 = nn.Conv3d(1, 1, 1) self.bn1d2 = nn.BatchNorm1d(1) self.bn2d2 = nn.BatchNorm2d(1) self.bn3d2 = nn.BatchNorm3d(1) self.relu = nn.ReLU() def forward(self, x): x = self.conv1d(x) x = self.bn1d(x) x = self.conv2d(x) x = self.bn2d(x) x = self.conv3d(x) x = self.bn3d(x) x = self.conv1d2(x) x = self.bn1d2(x) x = self.relu(x) x = self.conv2d2(x) x = self.bn2d2(x) x = self.relu(x) x = self.conv3d2(x) x = self.bn3d2(x) x = self.relu(x) return x # test train mode m = M().train() # currently we don't check if the module are configured with qconfig before fusion # TODO: if we decide to do that in the future, this test needs to # be updated # train mode fuse_fx is called in prepare_qat_fx m = prepare_qat_fx(m, {}) expected_nodes = [ ns.call_module(nni.ConvBn1d), ns.call_module(nni.ConvBn2d), ns.call_module(nni.ConvBn3d), ns.call_module(nni.ConvBnReLU1d), ns.call_module(nni.ConvBnReLU2d), ns.call_module(nni.ConvBnReLU3d), ] expected_occurrence = { ns.call_module(nn.ReLU): 0 } self.checkGraphModuleNodes( m, expected_node_list=expected_nodes, expected_node_occurrence=expected_occurrence) # test eval mode m = M().eval() from torch.quantization.quantize_fx import fuse_fx # fuse_fx is a top level api and only supports eval mode m = fuse_fx(m) expected_nodes = [ ns.call_module(nn.Conv1d), ns.call_module(nn.Conv2d), ns.call_module(nn.Conv3d), ns.call_module(nni.ConvReLU1d), ns.call_module(nni.ConvReLU2d), ns.call_module(nni.ConvReLU3d), ] # ConvBnRelu1d is not fused expected_occurrence = { ns.call_module(nn.ReLU): 0 } self.checkGraphModuleNodes( m, expected_node_list=expected_nodes, expected_node_occurrence=expected_occurrence) def test_fuse_module_relu(self): class M(torch.nn.Module): def __init__(self): super().__init__() self.conv1d = nn.Conv1d(1, 1, 1) self.conv2d = nn.Conv2d(1, 1, 1) self.conv3d = nn.Conv3d(1, 1, 1) self.bn1d = nn.BatchNorm1d(1) self.bn2d = nn.BatchNorm2d(1) self.bn3d = nn.BatchNorm3d(1) self.relu = nn.ReLU() def forward(self, x): x = self.conv1d(x) x = self.relu(x) x = self.conv2d(x) x = self.relu(x) x = self.conv3d(x) x = self.relu(x) x = self.bn1d(x) x = self.relu(x) x = self.bn2d(x) x = self.relu(x) x = self.bn3d(x) x = self.relu(x) return x m = M().eval() from torch.quantization.quantize_fx import fuse_fx m = fuse_fx(m) expected_nodes = [ ns.call_module(nni.ConvReLU1d), ns.call_module(nni.ConvReLU2d), ns.call_module(nni.ConvReLU3d), ns.call_module(nni.BNReLU2d), ns.call_module(nni.BNReLU3d), ] self.checkGraphModuleNodes(m, expected_node_list=expected_nodes) @skipIfNoFBGEMM class TestQuantizeFx(QuantizationTestCase): def test_pattern_match(self): """ test MatchAllNode with conv - bn - add - relu pattern """ class M(torch.nn.Module): def __init__(self): super().__init__() self.conv = nn.Conv2d(1, 1, 1) self.bn = nn.BatchNorm2d(1) self.relu = nn.ReLU() def forward(self, x, y): x = self.conv(x) x = self.bn(x) x = x + y x = self.relu(x) return x pattern = (nn.ReLU, (operator.add, (nn.BatchNorm2d, nn.Conv2d), MatchAllNode)) m = torch.fx.symbolic_trace(M()) modules = dict(m.named_modules()) for n in m.graph.nodes: if n.op == 'call_module' and type(modules[n.target]) == nn.ReLU: self.assertTrue(is_match(modules, n, pattern)) def _get_conv_linear_test_cases(self): ''' Returns a list of test cases, with format: is_dynamic, ModuleClass, module_constructor_inputs, inputs, quantized_node, weight_prepack_op ''' class Conv(torch.nn.Module): def __init__(self, weight): super().__init__() self.weight = torch.nn.Parameter(weight) self.stride = (1, 1) self.padding = (0, 0) self.dilation = (1, 1) self.groups = 1 def forward(self, x): return F.conv2d(x, self.weight, None, self.stride, self.padding, self.dilation, self.groups) conv_input = torch.rand(1, 3, 224, 224) conv_weight = torch.rand(3, 3, 3, 3) class Linear(torch.nn.Module): def __init__(self, weight): super().__init__() self.weight = torch.nn.Parameter(weight) def forward(self, x): return F.linear(x, self.weight) linear_input = torch.rand(8, 5) linear_weight = torch.rand(10, 5) class LinearModule(torch.nn.Module): def __init__(self): super().__init__() self.linear = torch.nn.Linear(5, 10) def forward(self, x): return self.linear(x) linear_module_input = torch.rand(8, 5) tests = [ (False, Conv, (conv_weight,), (conv_input,), ns.call_function(torch.ops.quantized.conv2d), ns.call_function(torch.ops.quantized.conv2d_prepack)), (True, Linear, (linear_weight,), (linear_input,), ns.call_function(torch.ops.quantized.linear_dynamic), ns.call_function(torch.ops.quantized.linear_prepack)), (False, Linear, (linear_weight,), (linear_input,), ns.call_function(torch.ops.quantized.linear), ns.call_function(torch.ops.quantized.linear_prepack)), (True, LinearModule, (), (linear_module_input,), ns.call_module(nnqd.Linear), None), (False, LinearModule, (), (linear_module_input,), ns.call_module(nnq.Linear), None), ] return tests """ Unit tests for functionalities """ @skipIfNoFBGEMM def test_functional_no_debug(self): """ Test quantizing functional conv and linear """ tests = self._get_conv_linear_test_cases() for (is_dynamic, ModuleClass, module_constructor_inputs, inputs, quantized_node, weight_prepack_node) in tests: quant_type = QuantType.DYNAMIC if is_dynamic else QuantType.STATIC node_occurrence = dict() if weight_prepack_node: node_occurrence[weight_prepack_node] = 0 self.checkGraphModeFxOp( ModuleClass(*module_constructor_inputs), inputs, quant_type, expected_node=quantized_node, expected_node_occurrence=node_occurrence, debug=False) @skipIfNoFBGEMM def test_functional_debug(self): """ Test quantizing functional conv and linear with debug option """ tests = self._get_conv_linear_test_cases() for (is_dynamic, ModuleClass, module_constructor_inputs, inputs, quantized_node, weight_prepack_node) in tests: quant_type = QuantType.DYNAMIC if is_dynamic else QuantType.STATIC node_occurrence = dict() if weight_prepack_node: node_occurrence[weight_prepack_node] = 0 node_occurrence[quantized_node] = 0 self.checkGraphModeFxOp( ModuleClass(*module_constructor_inputs), inputs, quant_type, expected_node_occurrence=node_occurrence, debug=True) @skipIfNoFBGEMM def test_dynamic_quant_weight_observer(self): ''' Test that weight observer is run in convert step ''' class M(torch.nn.Module): def __init__(self, weight): super().__init__() self.weight = torch.nn.Parameter(weight) def forward(self, x): return F.linear(x, self.weight) m = M(torch.rand(1, 1)).eval() qconfig = default_dynamic_qconfig qconfig_dict = {'': qconfig} prepared = prepare_fx(m, qconfig_dict) quantized = convert_fx(prepared, debug=True) qparams = (quantized._input_scale_0, quantized._input_zero_point_0) weight_obs = qconfig.weight() weight_obs(quantized.weight) # Get the actual value to avoid tensor size mismatch error, torch.Size([]) vs torch.Size([1]) ref_qparams = (weight_obs.calculate_qparams()[0].item(), weight_obs.calculate_qparams()[1].item()) self.assertEqual(qparams, ref_qparams) def test_conv_bn_relu(self): convs = { 1: nn.Conv1d, 2: nn.Conv2d, 3: nn.Conv3d, } bns = { 1: nn.BatchNorm1d, 2: nn.BatchNorm2d, 3: nn.BatchNorm3d, } quantized_convs = { 1: nnq.Conv1d, 2: nnq.Conv2d, 3: nnq.Conv3d, } quantized_conv_relus = { 1: nniq.ConvReLU1d, 2: nniq.ConvReLU2d, 3: nniq.ConvReLU3d, } class M(torch.nn.Module): def __init__(self, dim, has_relu): super().__init__() self.conv = convs[dim](3, 3, 3) self.bn = bns[dim](3) self.relu = nn.ReLU() if has_relu else nn.Identity() self.has_relu = has_relu self.quant = QuantStub() self.dequant = DeQuantStub() def forward(self, x): x = self.quant(x) x = self.conv(x) x = self.bn(x) if self.has_relu: x = self.relu(x) x = self.dequant(x) return x options = itertools.product([1, 2], [True, False], self.static_quant_types) for dim, has_relu, quant_type in options: expected_node = ns.call_module( quantized_conv_relus[dim] if has_relu else quantized_convs[dim]) m = M(dim, has_relu) m_eager = copy.deepcopy(m) result = self.checkGraphModeFxOp( m, self.img_data_dict[dim], quant_type, expected_node=expected_node, ) # check numerics qengine = torch.backends.quantized.engine if quant_type == QuantType.STATIC: m_eager.eval() qconfig = get_default_qconfig(qengine) prepare_fn = prepare else: m_eager.train() qconfig = get_default_qat_qconfig(qengine) prepare_fn = prepare_qat fuse_list = ["conv", "bn"] if has_relu: fuse_list.append("relu") fuse_modules(m_eager, fuse_list, inplace=True) m_eager.qconfig = qconfig m_eager = prepare_fn(m_eager) m_eager(*self.img_data_dict[dim][0]) m_eager = convert(m_eager) result_eager = m_eager(*self.img_data_dict[dim][0]) self.assertEqual(result, result_eager) @skipIfNoFBGEMM def test_dynamic_quant_fp16(self): class Linear(torch.nn.Module): def __init__(self, weight): super().__init__() self.weight = torch.nn.Parameter(weight) def forward(self, x): return F.linear(x, self.weight) linear_input = torch.rand(8, 5) linear_weight = torch.rand(10, 5) class LinearModule(torch.nn.Module): def __init__(self): super().__init__() self.linear = torch.nn.Linear(5, 10) def forward(self, x): return self.linear(x) linear_module_input = torch.rand(8, 5) tests = [ (Linear, (linear_weight,), (linear_input,), ns.call_function(torch.ops.quantized.linear_dynamic), ns.call_function(torch.ops.quantized.linear_prepack_fp16)), (LinearModule, (), (linear_module_input,), ns.call_module(nnqd.Linear), None), ] for (ModuleClass, module_constructor_inputs, inputs, quantized_node, weight_prepack_node) in tests: for debug in [True, False]: node_occurrence = dict() if weight_prepack_node: node_occurrence[weight_prepack_node] = 0 m = ModuleClass(*module_constructor_inputs).eval() qconfig_dict = {"": float16_dynamic_qconfig} m = prepare_fx(m, qconfig_dict) m = convert_fx(m, debug=debug) self.checkGraphModuleNodes(m, expected_node_occurrence=node_occurrence) @unittest.skipIf(not TEST_MULTIGPU, "multi-GPU not supported") @unittest.skipIf(not TEST_CUDA, "CUDA unavailable") @override_qengines def test_qat_prepare_device_affinity(self): """ Tests that FX QAT prepare pass respects device affinity """ class Model(nn.Module): def __init__(self): super(Model, self).__init__() self.conv = nn.Conv2d(1, 1, 1) self.bn = nn.BatchNorm2d(1) self.relu = nn.ReLU() def forward(self, x): x = self.conv(x) x = self.bn(x) x = self.relu(x) return x model = Model() qengine = torch.backends.quantized.engine qconfig_dict = {'': torch.quantization.get_default_qat_qconfig(qengine)} device = torch.device('cuda:0') model.to(device) # QAT prepare model = prepare_qat_fx(model, qconfig_dict) # ensure that running an input on CUDA works without any needed changes input = torch.randn(4, 1, 4, 4, device=device) model(input) # ensure all buffers and parameters are on the device we expect model_devices = {p.device for p in model.parameters()} | \ {p.device for p in model.buffers()} self.assertEqual(len(model_devices), 1) model_device = next(iter(model_devices)) self.assertEqual(model_device, device) @skipIfNoFBGEMM def test_dict_output(self): """ Make sure quantization runs for models with dictionary output """ class M(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(1, 1, 1) def forward(self, x): return {"output": self.conv(x["input"])} dict_input = {"input": torch.randn(1, 1, 1, 1)} m = M().eval() qconfig_dict = {"": default_qconfig} m = prepare_fx(m, qconfig_dict) m(dict_input) m = convert_fx(m) m(dict_input) @override_qengines def test_attention(self): """ Make sure quantization runs for a corner case in attention module """ class M(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(1, 1, 1) def forward(self, x): x = self.conv(x) q, k, v = x.chunk(3, dim=0) q = q.contiguous().view(-1, 1).transpose(0, 1) k = k.contiguous().view(-1, 1).transpose(0, 1) v = v.contiguous().view(-1, 1).transpose(0, 1) torch._assert( k.size(1) == 1, "key size should be equal to 1" ) r = torch.mm(k, v) return q * k + r tensor_input = torch.randn(3, 1, 1, 1) m = M().eval() qconfig_dict = { "": None, "object_type": [ (nn.Conv2d, default_qconfig), ] } # make sure it runs m = prepare_fx(m, qconfig_dict) m(tensor_input) m = convert_fx(m) m(tensor_input) def _test_standalone_module( self, interface_config, prepare_count_check, standalone_prepare_count_check, convert_count_check, standalone_convert_count_check): """ Test standalone module with different quantized input/quantized output configurations """ class StandaloneModule(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(1, 1, 1) def forward(self, x): return self.conv(x) class M(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(1, 1, 1) self.standalone = StandaloneModule() def forward(self, x): x = self.conv(x) x = self.standalone(x) return x class RefM(torch.nn.Module): def __init__(self): super().__init__() self.conv1 = torch.nn.Conv2d(1, 1, 1) self.conv2 = torch.nn.Conv2d(1, 1, 1) def forward(self, x): x = self.conv1(x) x = self.conv2(x) return x data = torch.randn(1, 1, 1, 1) # instantiate M and RefM and align the parameters original_m = M().eval() original_ref_m = RefM().eval() original_ref_m.conv1.weight = torch.nn.Parameter(original_m.conv.weight.detach()) original_ref_m.conv1.bias = torch.nn.Parameter(original_m.conv.bias.detach()) original_ref_m.conv2.weight = torch.nn.Parameter(original_m.standalone.conv.weight.detach()) original_ref_m.conv2.bias = torch.nn.Parameter(original_m.standalone.conv.bias.detach()) for is_name in [True, False]: if is_name: prepare_config = { "standalone_module_name": [("standalone", None, interface_config)] } else: prepare_config = { "standalone_module_class": [(StandaloneModule, None, interface_config)] } original_m_copy = copy.deepcopy(original_m) original_ref_m_copy = copy.deepcopy(original_ref_m) qconfig_dict = {"": default_qconfig} # check prepared model m = prepare_fx( original_m_copy, qconfig_dict, prepare_custom_config_dict=prepare_config) # calibration m(data) self.checkGraphModuleNodes(m, expected_node_occurrence=prepare_count_check) self.checkGraphModuleNodes(m.standalone, expected_node_occurrence=standalone_prepare_count_check) # check converted/quantized model m = convert_fx(m) self.checkGraphModuleNodes(m, expected_node_occurrence=convert_count_check) self.checkGraphModuleNodes(m.standalone, expected_node_occurrence=standalone_convert_count_check) res = m(data) # quantize the reference model ref_m = prepare_fx(original_ref_m_copy, qconfig_dict) ref_m(data) ref_m = convert_fx(ref_m) ref_res = ref_m(data) self.assertEqual(res, ref_res) def test_standalone_module_float_interface(self): float_interface_config = { "input_quantized_idxs": [], # float input "output_quantized_idxs": [], # float output } interface_config = float_interface_config # input and output of first conv, observer for standalone module # will be inserted in the standalone module itself prepare_count_check = { ns.call_module(torch.quantization.MinMaxObserver): 2 } # for input and output of conv in the standalone module standalone_prepare_count_check = { ns.call_module(torch.quantization.MinMaxObserver): 2 } convert_count_check = { ns.call_function(torch.quantize_per_tensor) : 1, ns.call_module(nnq.Conv2d) : 1, ns.call_method("dequantize") : 1, } standalone_convert_count_check = { # standalone module will take float as input and output # so we'll see quantize and dequantize in the modoule ns.call_function(torch.quantize_per_tensor) : 1, ns.call_module(nnq.Conv2d): 1, ns.call_method("dequantize") : 1, } self._test_standalone_module( interface_config, prepare_count_check, standalone_prepare_count_check, convert_count_check, standalone_convert_count_check) def test_standalone_module_quantized_interface(self): quantized_interface_config = { "input_quantized_idxs": [0], # quantized input "output_quantized_idxs": [0], # quantized output } interface_config = quantized_interface_config # observer for input and output of first conv prepare_count_check = { ns.call_module(torch.quantization.MinMaxObserver): 2 } # for output of conv in the standalone module standalone_prepare_count_check = { ns.call_module(torch.quantization.MinMaxObserver): 1 } convert_count_check = { # quantizing input for conv ns.call_function(torch.quantize_per_tensor) : 1, ns.call_module(nnq.Conv2d) : 1, # dequantizing output of standalone module ns.call_method("dequantize") : 1, } standalone_convert_count_check = { # quantization of input happens in parent module # quantization of output happens in the quantized conv module ns.call_function(torch.quantize_per_tensor) : 0, ns.call_module(nnq.Conv2d): 1, # dequantization for output happens in parent module ns.call_method("dequantize") : 0, } self._test_standalone_module( interface_config, prepare_count_check, standalone_prepare_count_check, convert_count_check, standalone_convert_count_check) @skipIfNoFBGEMM def test_qconfig_none(self): class M(torch.nn.Module): def __init__(self): super(M, self).__init__() self.conv1 = nn.Conv2d(1, 1, 1) self.conv2 = nn.Conv2d(1, 1, 1) def forward(self, x): x = self.conv1(x) x = self.conv2(x) return x m = M().eval() qconfig_dict = {"": default_qconfig, "module_name": [("conv2", None)]} m = prepare_fx(m, qconfig_dict) data = torch.randn(1, 1, 1, 1) m(data) m = convert_fx(m) m(data) # first conv is quantized, second conv is not quantized node_list = [ ns.call_function(torch.quantize_per_tensor), ns.call_module(nnq.Conv2d), ns.call_method("dequantize"), ns.call_module(nn.Conv2d), ] self.checkGraphModuleNodes(m, expected_node_list=node_list) def test_qconfig_module_type(self): class M(torch.nn.Module): def __init__(self): super(M, self).__init__() self.conv1 = nn.Conv2d(1, 1, 1) self.conv2 = nn.Conv2d(1, 1, 1) def forward(self, x): x = self.conv1(x) x = self.conv2(x) return x m = M().eval() qconfig_dict = {"object_type": [(torch.nn.Conv2d, default_qconfig)]} m = prepare_fx(m, qconfig_dict) data = torch.randn(1, 1, 1, 1) m(data) m = convert_fx(m) m(data) # first conv is quantized, second conv is not quantized node_list = [ ns.call_function(torch.quantize_per_tensor), ns.call_module(nnq.Conv2d), ns.call_module(nnq.Conv2d), ns.call_method("dequantize"), ] self.checkGraphModuleNodes(m, expected_node_list=node_list) def test_qconfig_function(self): class M(torch.nn.Module): def __init__(self): super(M, self).__init__() def forward(self, x, y): return x + y m = M().eval() qconfig_dict = {"object_type": [(operator.add, default_qconfig)]} m = prepare_fx(m, qconfig_dict) data = torch.randn(1, 1, 1, 1) m(data, data) m = convert_fx(m) m(data, data) # first conv is quantized, second conv is not quantized node_list = [ ns.call_function(torch.quantize_per_tensor), ns.call_function(torch.ops.quantized.add), ns.call_method("dequantize"), ] self.checkGraphModuleNodes(m, expected_node_list=node_list) def test_qconfig_module_name_regex(self): class M(torch.nn.Module): def __init__(self): super(M, self).__init__() self.conv1 = nn.Conv2d(1, 1, 1) self.conv2 = nn.Conv2d(1, 1, 1) def forward(self, x): x = self.conv1(x) x = self.conv2(x) return x m = M().eval() qconfig_dict = {"module_name_regex": [("conv*", default_qconfig)]} m = prepare_fx(m, qconfig_dict) data = torch.randn(1, 1, 1, 1) m(data) m = convert_fx(m) m(data) # first conv is quantized, second conv is not quantized node_list = [ ns.call_function(torch.quantize_per_tensor), ns.call_module(nnq.Conv2d), ns.call_module(nnq.Conv2d), ns.call_method("dequantize"), ] self.checkGraphModuleNodes(m, expected_node_list=node_list) def test_qconfig_precedence(self): class M(torch.nn.Module): def __init__(self): super(M, self).__init__() self.linear = nn.Linear(1, 1) self.conv = nn.Conv2d(1, 1, 1) self.module_conv1 = nn.Conv2d(1, 1, 1) self.module_conv2 = nn.Conv2d(1, 1, 1) def forward(self, x): # global x = self.linear(x) # global + object_type --> object_type x = self.conv(x) # global + object_type + module_name_regex --> module_name_regex x = self.module_conv1(x) # global + object_type + module_name_regex + module_name --> module_name x = self.module_conv2(x) return x m = M().eval() global_qconfig = default_qconfig object_type_qconfig = default_dynamic_qconfig module_name_regex_qconfig = float16_dynamic_qconfig module_name_qconfig = default_qat_qconfig qconfig_dict = { "": global_qconfig, "object_type": [(nn.Conv2d, object_type_qconfig)], "module_name_regex": [("module_conv*", module_name_regex_qconfig)], "module_name": [("module_conv2", module_name_qconfig)]} m = prepare_fx(m, qconfig_dict) self.assertEqual(m.linear.qconfig, global_qconfig) self.assertEqual(m.conv.qconfig, object_type_qconfig) self.assertEqual(m.module_conv1.qconfig, module_name_regex_qconfig) self.assertEqual(m.module_conv2.qconfig, module_name_qconfig) def test_remove_qconfig(self): class M(torch.nn.Module): def __init__(self): super().__init__() self.avg_pool = torch.nn.AvgPool2d(1) def forward(self, x): return self.avg_pool(x) m = M().eval() qconfig_dict = {'': default_qconfig} m = prepare_fx(m, qconfig_dict) data = torch.randn(1, 1, 1, 1) m(data) m = convert_fx(m) m(data) for name, module in m.named_modules(): self.assertFalse(hasattr(module, 'qconfig'), 'qconfig is not removed for ' + name) def test_default_quant_after_none_qconfig(self): """ Make sure default quant is inserted properly""" class M(torch.nn.Module): def __init__(self): super().__init__() self.conv1 = torch.nn.Conv2d(1, 1, 1) self.conv2 = torch.nn.Conv2d(1, 1, 1) def forward(self, x): x = self.conv1(x) x = x.transpose(1, 2) x = self.conv2(x) m = M().eval() qconfig_dict = { "": default_qconfig, "module_name": [ ("conv1", None) ] } m = prepare_fx(m, qconfig_dict) m = convert_fx(m) def test_qconfig_for_call_method(self): class Sub(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(1, 1, 1) def forward(self, x): x = x.transpose(2, 3) x = self.conv(x) return x.transpose(2, 3) class M(torch.nn.Module): def __init__(self): super().__init__() self.sub = Sub() self.conv1 = torch.nn.Conv2d(1, 1, 1) self.conv2 = torch.nn.Conv2d(1, 1, 1) def forward(self, x): x = self.conv1(x) x = self.sub(x) x = self.conv2(x) return x.transpose(2, 3) qconfig_dict1 = {"": default_qconfig, "module_name": [("sub", None)]} # since sub is configured to have qconfig None, we should dequantize the output # of self.conv1 and quantize the input of self.conv2 # dequantize after conv2 should happen after transpose since # it is configured with default_qconfig # nodes in Sub module instance is not quantized node_list1 = [ ns.call_function(torch.quantize_per_tensor), ns.call_module(nnq.Conv2d), ns.call_method("dequantize"), ns.call_method("transpose"), ns.call_module(nn.Conv2d), ns.call_method("transpose"), ns.call_function(torch.quantize_per_tensor), ns.call_module(nnq.Conv2d), ns.call_method("transpose"), ns.call_method("dequantize") ] qconfig_dict2 = {"": None, "module_name": [("sub", default_qconfig)]} # Only nodes in Sub module instance are quantized # the first transpose is not quantized because the input is not quantized node_list2 = [ ns.call_module(nn.Conv2d), ns.call_method("transpose"), ns.call_function(torch.quantize_per_tensor), ns.call_module(nnq.Conv2d), ns.call_method("transpose"), ns.call_method("dequantize"), ns.call_module(nn.Conv2d), ns.call_method("transpose"), ] for qconfig_dict, node_list in [ (qconfig_dict1, node_list1), (qconfig_dict2, node_list2) ]: m = M().eval() m = prepare_fx(m, qconfig_dict) m(torch.randn(2, 1, 3, 3)) m = convert_fx(m) self.checkGraphModuleNodes(m, expected_node_list=node_list) # make sure it runs m(torch.randn(2, 1, 3, 3)) def test_qconfig_for_call_func(self): class Linear(torch.nn.Module): def __init__(self): super().__init__() self.w = torch.ones(5, 5) self.b = torch.zeros(5) def forward(self, x): return torch.nn.functional.linear(x, self.w, self.b) class M(torch.nn.Module): def __init__(self): super().__init__() self.mods1 = torch.nn.Sequential( Linear(), Linear() ) self.mods2 = Linear() def forward(self, x): x = self.mods1(x) x = self.mods2(x) return x model = M().eval() qconfig_dict = {"": default_qconfig, "module_name": [("mods2", None)]} m = prepare_fx(model, qconfig_dict) m(torch.rand(5, 5)) m = convert_fx(m) node_list = [ ns.call_function(torch.quantize_per_tensor), ns.call_function(torch.ops.quantized.linear), ns.call_function(torch.ops.quantized.linear), ns.call_method('dequantize'), ns.call_function(torch.nn.functional.linear) ] self.checkGraphModuleNodes(m, expected_node_list=node_list) m(torch.rand(5, 5)) def test_preserve_attributes(self): class M(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(1, 1, 1) def forward(self, x): return self.conv(x) m = M() m.eval() m.preserved_attr = 3 prepare_custom_config_dict = { "preserved_attributes": ["preserved_attr"] } m = prepare_fx(m, {"": default_qconfig}, prepare_custom_config_dict) def assertAttrPreserved(m): self.assertTrue(hasattr(m, "preserved_attr")) self.assertTrue(m.preserved_attr, 3) assertAttrPreserved(m) convert_custom_config_dict = { "preserved_attributes": ["preserved_attr"] } m = convert_fx(m, convert_custom_config_dict=convert_custom_config_dict) assertAttrPreserved(m) @skipIfNoFBGEMM def test_qat_and_script(self): model = LinearModelWithSubmodule().train() qengine = torch.backends.quantized.engine qconfig_dict = {'': torch.quantization.get_default_qat_qconfig(qengine)} model = prepare_qat_fx(model, qconfig_dict) # ensure scripting works scripted = torch.jit.script(model) # run one round to make sure model runs x = torch.randn(5, 5) scripted(x) FileCheck().check_count('FakeQuantize = prim::GetAttr[name="', 4, exactly=True) \ .run(scripted.graph) # disable fake_quant and observer for epoch in range(3): if epoch == 1: scripted.apply(torch.quantization.disable_observer) if epoch == 2: scripted.apply(torch.quantization.disable_fake_quant) # ensure the fake_quant and observer have been disabled. matches = ['.fake_quant_enabled', '.observer_enabled'] for key, v in scripted.state_dict().items(): if any(x in key for x in matches): self.assertEqual(v, torch.tensor([0], dtype=torch.uint8)) # enable them back scripted.apply(torch.quantization.enable_fake_quant) scripted.apply(torch.quantization.enable_observer) for key, v in scripted.state_dict().items(): if any(x in key for x in matches): self.assertEqual(v, torch.tensor([1], dtype=torch.uint8)) @skipIfNoFBGEMM def test_save_observer_state_dict(self): orig = LinearModelWithSubmodule().eval() model = orig qconfig_dict = {'': torch.quantization.get_default_qconfig('fbgemm')} model = prepare_fx(model, qconfig_dict) # run it through input x = torch.randn(5, 5) model(x) quant = convert_fx(model) # save state_dict of model obs_dict = torch.quantization.get_observer_state_dict(model) b = io.BytesIO() torch.save(obs_dict, b) b.seek(0) # Load the stats into new model model_2 = orig model_2 = prepare_fx(model_2, qconfig_dict) loaded_dict = torch.load(b) torch.quantization.load_observer_state_dict(model_2, loaded_dict) quant_2 = convert_fx(model_2) # Verify that loaded state dict produces same results. self.assertEqual(quant(x), quant_2(x)) @skipIfNoFBGEMM def test_custom_module_class(self): class CustomModule(torch.nn.Module): def __init__(self): super().__init__() self.linear = torch.nn.Linear(3, 3) def forward(self, x): return self.linear(x) class ObservedCustomModule(torch.nn.Module): def __init__(self, linear): super().__init__() self.linear = linear def forward(self, x): return self.linear(x) @classmethod def from_float(cls, float_module): assert hasattr(float_module, 'qconfig') observed = cls(float_module.linear) observed.qconfig = float_module.qconfig return observed class StaticQuantCustomModule(torch.nn.Module): def __init__(self, linear): super().__init__() self.linear = linear def forward(self, x): return self.linear(x) @classmethod def from_observed(cls, observed_module): assert hasattr(observed_module, 'qconfig') assert hasattr(observed_module, 'activation_post_process') observed_module.linear.activation_post_process = \ observed_module.activation_post_process quantized = cls(nnq.Linear.from_float(observed_module.linear)) return quantized class DynamicQuantCustomModule(torch.nn.Module): def __init__(self, linear): super().__init__() self.linear = linear def forward(self, x): return self.linear(x) @classmethod def from_observed(cls, observed_module): assert hasattr(observed_module, 'qconfig') quantized = cls(nnqd.Linear.from_float(observed_module.linear)) return quantized class M(torch.nn.Module): def __init__(self): super().__init__() self.linear = torch.nn.Linear(3, 3) self.custom = CustomModule() def forward(self, x): x = self.linear(x) x = self.custom(x) return x class RefM(torch.nn.Module): def __init__(self): super().__init__() self.linear1 = torch.nn.Linear(3, 3) self.linear2 = torch.nn.Linear(3, 3) def forward(self, x): x = self.linear1(x) x = self.linear2(x) return x data = torch.randn(3, 3) # instantiate M and RefM and align the parameters original_m = M().eval() original_ref_m = RefM().eval() original_ref_m.linear1.weight = torch.nn.Parameter(original_m.linear.weight.detach()) original_ref_m.linear1.bias = torch.nn.Parameter(original_m.linear.bias.detach()) original_ref_m.linear2.weight = torch.nn.Parameter(original_m.custom.linear.weight.detach()) original_ref_m.linear2.bias = torch.nn.Parameter(original_m.custom.linear.bias.detach()) test_configs = { "static": (default_qconfig, StaticQuantCustomModule, 3), "dynamic": (default_dynamic_qconfig, DynamicQuantCustomModule, 0) } for quant_type in [QuantType.DYNAMIC]: key = quant_type_to_str(quant_type) qconfig, quantized_module_class, num_observers = test_configs[key] qconfig_dict = {"": qconfig} if key == "static": prepare_custom_config_dict = { "float_to_observed_custom_module_class": { "static": { CustomModule: ObservedCustomModule } } } convert_custom_config_dict = { "observed_to_quantized_custom_module_class": { "static": { ObservedCustomModule: quantized_module_class } } } else: prepare_custom_config_dict = { "non_traceable_module_class": [ CustomModule ] } convert_custom_config_dict = { "observed_to_quantized_custom_module_class": { "dynamic": { CustomModule: quantized_module_class } } } # check prepared model m = prepare_fx( original_m, qconfig_dict, prepare_custom_config_dict=prepare_custom_config_dict) # calibration m(data) # all activation observers are inserted in the top level module count_check = { ns.call_module(torch.quantization.MinMaxObserver): num_observers } self.checkGraphModuleNodes(m, expected_node_occurrence=count_check) # check converted/quantized model m = convert_fx( m, convert_custom_config_dict=convert_custom_config_dict) if quant_type == QuantType.STATIC: count_check = { ns.call_function(torch.quantize_per_tensor) : 1, ns.call_module(nnq.Linear) : 1, ns.call_method('dequantize') : 1, } self.checkGraphModuleNodes(m, expected_node_occurrence=count_check) self.assertEqual(type(m.custom), quantized_module_class) res = m(data) # quantize the reference model ref_m = prepare_fx(original_ref_m, qconfig_dict) ref_m(data) ref_m = convert_fx(ref_m) ref_res = ref_m(data) self.assertEqual(res, ref_res) @skipIfNoFBGEMM def test_non_traceable_module(self): class NonTraceable(torch.nn.Module): def __init__(self): super().__init__() def forward(self, x): for k in x.keys(): print(x[k]) return x class NonTraceable2(torch.nn.Module): def __init__(self): super().__init__() def forward(self, x): # data dependent control flow is not traceable for i in x: print(i) return x class M(torch.nn.Module): def __init__(self): super().__init__() self.m1 = NonTraceable() self.m2 = NonTraceable2() def forward(self, x): x = self.m1(x) x = self.m2(x) return x m = M().eval() qconfig_dict = {"": default_qconfig} prepare_custom_config_dict = { "non_traceable_module_name": [ "m1" ], "non_traceable_module_class": [ NonTraceable2 ] } m = prepare_fx( m, qconfig_dict, prepare_custom_config_dict=prepare_custom_config_dict) node_occurrence = { ns.call_module(NonTraceable) : 1, ns.call_module(NonTraceable2) : 1, } # make sure these modules are not traced self.checkGraphModuleNodes(m, expected_node_occurrence=node_occurrence) def test_prepared_model_deepcopy(self): """Ensures that copy.deepcopy works correctly on a prepared model. """ class M(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(1, 1, 1) self._foobar = 'foobar' self.foobar2 = 'foobar2' def forward(self, x): x = self.conv(x) return x m = M() m.eval() qconfig_dict = {'': torch.quantization.default_qconfig} prepared = prepare_fx(m, qconfig_dict) # calibrate prepared(torch.randn(4, 1, 4, 4)) # copy prepared_copy = copy.deepcopy(prepared) # quantize, should run with no errors quantized = convert_fx(prepared_copy) def test_dequantize(self): r""" Test to make sure dequantize node are placed before non-quantizable node """ class M(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(1, 1, 1) self.act = torch.nn.GELU() def forward(self, x): x = self.conv(x) return self.act(x) data = torch.rand(5, 1, 3, 3, dtype=torch.float) for quant_type in self.static_quant_types: node_list = [ ns.call_module(nnq.Conv2d), ns.call_method("dequantize"), ns.call_module(nn.GELU), ] self.checkGraphModeFxOp( M().eval(), (data,), quant_type, expected_node_list=node_list) def test_sequential(self): class M(torch.nn.Module): def __init__(self): super().__init__() self.convs = torch.nn.Sequential( torch.nn.Conv2d(1, 1, 1), torch.nn.Conv2d(1, 1, 1) ) def forward(self, x): x = self.convs(x) return x data = torch.rand(5, 1, 3, 3, dtype=torch.float) for quant_type in self.static_quant_types: node_list = [ ns.call_module(nnq.Conv2d), ns.call_module(nnq.Conv2d), ] self.checkGraphModeFxOp( M().eval(), (data,), quant_type, expected_node_list=node_list) def _test_quantized_inputs_outputs( self, prepare_custom_config_dict, prepare_count_check, convert_count_check): """ Test the option to have inputs and outputs of the graph quantized """ class M(torch.nn.Module): def __init__(self): super().__init__() self.conv1 = torch.nn.Conv2d(1, 1, 1) self.conv2 = torch.nn.Conv2d(1, 1, 1) def forward(self, x): x = self.conv1(x) x = self.conv2(x) return x # quantized input, quantized output m = M() qconfig_dict = {'': torch.quantization.default_qconfig} m.eval() mp = torch.quantization.quantize_fx.prepare_fx( m, qconfig_dict, prepare_custom_config_dict=prepare_custom_config_dict) self.checkGraphModuleNodes(mp, expected_node_occurrence=prepare_count_check) mp(torch.randn(1, 1, 4, 4)) mq = torch.quantization.quantize_fx.convert_fx(mp) self.checkGraphModuleNodes(mq, expected_node_occurrence=convert_count_check) def test_quantized_input_quantized_output(self): prepare_custom_config_dict = { 'input_quantized_idxs': [0], 'output_quantized_idxs': [0]} prepare_count_check = { ns.call_module(torch.quantization.MinMaxObserver): 2, } convert_count_check = { ns.call_function(torch.quantize_per_tensor): 0, ns.call_method('dequantize'): 0, } self._test_quantized_inputs_outputs( prepare_custom_config_dict, prepare_count_check, convert_count_check) def test_fp32_input_quantized_output(self): prepare_custom_config_dict = { 'output_quantized_idxs': [0]} prepare_count_check = { ns.call_module(torch.quantization.MinMaxObserver): 3, } convert_count_check = { ns.call_function(torch.quantize_per_tensor): 1, ns.call_method('dequantize'): 0, } self._test_quantized_inputs_outputs( prepare_custom_config_dict, prepare_count_check, convert_count_check) def test_quantized_input_fp32_output(self): prepare_custom_config_dict = { 'input_quantized_idxs': [0]} prepare_count_check = { ns.call_module(torch.quantization.MinMaxObserver): 2, } convert_count_check = { ns.call_function(torch.quantize_per_tensor): 0, ns.call_method('dequantize'): 1, } self._test_quantized_inputs_outputs( prepare_custom_config_dict, prepare_count_check, convert_count_check) def test_fp32_input_fp32_output(self): prepare_custom_config_dict = {} prepare_count_check = { ns.call_module(torch.quantization.MinMaxObserver): 3, } convert_count_check = { ns.call_function(torch.quantize_per_tensor): 1, ns.call_method('dequantize'): 1, } self._test_quantized_inputs_outputs( prepare_custom_config_dict, prepare_count_check, convert_count_check) @skipIfNoFBGEMM def test_convtranspose_per_channel_fails_early(self): r""" Verifies that attempting to quantize a ConvTranspose module with per-Channel weight observers fails in the prepare step, as opposed to the convert step. """ m = torch.nn.Sequential(torch.nn.ConvTranspose2d(1, 1, 1)) m.eval() qconfig_dict = {'': torch.quantization.get_default_qconfig('fbgemm')} with self.assertRaises(AssertionError) as context: mp = prepare_fx(m, qconfig_dict) self.assertTrue( str(context.exception) == 'Per channel weight observer is not supported yet for ConvTranspose{n}d.') @skipIfNoFBGEMM def test_qparams_buffers(self): class Linear(torch.nn.Module): def __init__(self): super().__init__() self.w = torch.ones(5, 5) self.b = torch.zeros(5) def forward(self, x): return torch.nn.functional.linear(x, self.w, self.b) class M(torch.nn.Module): def __init__(self): super().__init__() self.mods1 = torch.nn.Sequential( Linear(), Linear() ) self.mods2 = Linear() def forward(self, x): x = self.mods1(x) x = self.mods2(x) return x model = M().eval() qconfig_dict = {"": default_qconfig} m = prepare_fx(model, qconfig_dict) m(torch.rand(5, 5)) m = convert_fx(m) keys = m.state_dict().keys() quant_scale_count = quant_zero_point = scale_count = zero_point_count = 0 for k in keys: if 'input_scale' in k: quant_scale_count = quant_scale_count + 1 elif 'input_zero_point' in k: quant_zero_point = quant_zero_point + 1 elif 'scale' in k: scale_count = scale_count + 1 elif 'zero_point' in k: zero_point_count = zero_point_count + 1 # Expect each quantized linear op to have a scale and zero point self.assertTrue(scale_count == 3, "Expect each quantized linear op to have a scale in state_dict") self.assertTrue(zero_point_count == 3, "Expect each quantized linear op to have a zero_point in state_dict") # ensure it runs m(torch.rand(5, 5)) # ensure it is scriptable scripted = torch.jit.script(m) scripted_keys = scripted.state_dict().keys() self.assertTrue(scripted_keys == keys, "Expected the scripted model to preserve the state_dict") assert hasattr(m, "mods1_0_input_scale_0") assert hasattr(m, "mods1_0_input_zero_point_0") assert hasattr(m, "mods1_0_scale_0") assert hasattr(m, "mods1_0_zero_point_0") assert hasattr(m, "mods1_1_scale_0") assert hasattr(m, "mods1_1_zero_point_0") assert hasattr(m, "mods2_scale_0") assert hasattr(m, "mods2_zero_point_0") @skipIfNoFBGEMM def test_packed_weight_fused_op(self): class Linear(torch.nn.Module): def __init__(self): super().__init__() self.w = torch.ones(5, 5) self.b = torch.zeros(5) def forward(self, x): return F.linear(x, self.w, self.b) class M(torch.nn.Module): def __init__(self): super().__init__() self.mods1 = torch.nn.Sequential( Linear(), Linear() ) self.mods2 = Linear() self.relu = F.relu def forward(self, x): x = self.mods1(x) x = self.mods2(x) x = self.relu(x) return x model = M().eval() qconfig_dict = {"": default_qconfig} m = prepare_fx(model, qconfig_dict) m(torch.rand(5, 5)) m = convert_fx(m) assert hasattr(m, "mods1_0_packed_weight_0") assert hasattr(m, "mods1_1_packed_weight_0") assert hasattr(m, "mods2_packed_weight_0") @skipIfNoFBGEMM class TestQuantizeFxOps(QuantizationTestCase): """Unit tests for individual ops """ @skipIfNoFBGEMM def test_linear_module(self): class ModuleLinear(torch.nn.Module): def __init__(self, has_relu=False, f_relu=False): super(ModuleLinear, self).__init__() self.linear = torch.nn.Linear(30, 4).float() if has_relu: if f_relu: self.relu = F.relu else: self.relu = torch.nn.ReLU() else: self.relu = torch.nn.Identity() def forward(self, x): return self.relu(self.linear(x)) data = (torch.rand((1, 30), dtype=torch.float),) options = itertools.product( [ModuleLinear(has_relu=False)], self.all_quant_types) quantized_nodes = { # quant_type: QuantType.DYNAMIC: ns.call_module(nnqd.Linear), QuantType.STATIC: ns.call_module(nnq.Linear), # note that we are checking the final result QuantType.QAT: ns.call_module(nnq.Linear), } for model, quant_type in options: self.checkGraphModeFxOp( model, data, quant_type, quantized_nodes[quant_type]) for f_relu, quant_type in itertools.product([True, False], [QuantType.STATIC, QuantType.QAT]): for model, quantized_node in [ (ModuleLinear(has_relu=True, f_relu=f_relu), ns.call_module(nniq.LinearReLU))]: self.checkGraphModeFxOp(model, data, quant_type, quantized_node) @skipIfNoFBGEMM def test_functional_linear(self): class FuncLinear(torch.nn.Module): def __init__(self, use_bias, has_relu, f_relu): super(FuncLinear, self).__init__() self.w = torch.randn(4, 30) self.b = torch.randn(4) self.use_bias = use_bias if has_relu: if f_relu: self.relu = F.relu else: self.relu = torch.nn.ReLU() else: self.relu = torch.nn.Identity() def forward(self, x): if self.use_bias: x = F.linear(x, self.w, self.b) else: x = F.linear(x, self.w) x = self.relu(x) return x data = (torch.rand((1, 30), dtype=torch.float),) quant_type_to_prepare_expected_node_occurrence = { QuantType.DYNAMIC: {}, # There should be 3 observers: after input, weight and activation. QuantType.STATIC: { ns.call_module(torch.quantization.HistogramObserver): 2, ns.call_module(torch.quantization.PerChannelMinMaxObserver): 1, }, # There should be 3 observers: after input, weight and activation. QuantType.QAT: { ns.call_module(torch.quantization.FakeQuantize): 3, }, } quant_type_to_qlinear_fun = { QuantType.DYNAMIC: ns.call_function(torch.ops.quantized.linear_dynamic), QuantType.STATIC: ns.call_function(torch.ops.quantized.linear), QuantType.QAT: ns.call_function(torch.ops.quantized.linear), } quant_type_to_qlinear_relu_fun = { # we don't have linear_relu_dynamic QuantType.DYNAMIC: ns.call_function(torch.ops.quantized.linear_dynamic), QuantType.STATIC: ns.call_function(torch.ops.quantized.linear_relu), QuantType.QAT: ns.call_function(torch.ops.quantized.linear_relu), } options = itertools.product( self.all_quant_types, (True, False), # use_bias (True, False), # has_relu (True, False), # functional relu ) for quant_type, use_bias, has_relu, f_relu in options: model = FuncLinear(use_bias, has_relu, f_relu) if has_relu: qlinear_fun = quant_type_to_qlinear_relu_fun[quant_type] else: qlinear_fun = quant_type_to_qlinear_fun[quant_type] convert_node_occurrence = { ns.call_function(torch.quantize_per_tensor): 1 if quant_type != QuantType.DYNAMIC else 0, qlinear_fun: 1, ns.call_method("dequantize"): 1 if quant_type != QuantType.DYNAMIC else 0 } prepare_expected_node_occurrence = \ quant_type_to_prepare_expected_node_occurrence[quant_type] self.checkGraphModeFxOp( model, data, quant_type, qlinear_fun, prepare_expected_node_occurrence=prepare_expected_node_occurrence, expected_node_occurrence=convert_node_occurrence) @skipIfNoFBGEMM def test_conv_module(self): conv_module = {1 : torch.nn.Conv1d, 2 : torch.nn.Conv2d, 3 : torch.nn.Conv3d} class ConvWrapper(torch.nn.Module): def __init__(self, dim): super(ConvWrapper, self).__init__() self.conv = conv_module[dim](3, 3, 3).float() def forward(self, x): return self.conv(x) options = itertools.product([1, 2, 3], self.static_quant_types) quantized_nodes = { # dim 1: ns.call_module(nnq.Conv1d), 2: ns.call_module(nnq.Conv2d), 3: ns.call_module(nnq.Conv3d), } for dim, quant_type in options: model = self.checkGraphModeFxOp( ConvWrapper(dim), self.img_data_dict[dim], quant_type, quantized_nodes[dim]) @skipIfNoFBGEMM def test_functional_conv(self): """ Test for function conv and functional conv + relu """ convs = { 1: torch.nn.functional.conv1d, 2: torch.nn.functional.conv2d, 3: torch.nn.functional.conv3d, } class FuncConv(torch.nn.Module): def __init__(self, dim, use_bias, has_relu, f_relu): super().__init__() self.dim = dim self.w = torch.randn(tuple([3] * (dim + 2))) self.b = torch.randn(3) if use_bias else None self.stride = tuple([1] * dim) self.padding = tuple([0] * dim) self.dilation = tuple([1] * dim) self.groups = 1 self.use_bias = use_bias if has_relu: if f_relu: self.relu = F.relu else: self.relu = torch.nn.ReLU() else: self.relu = torch.nn.Identity() def forward(self, x): x = convs[self.dim](x, self.w, self.b, self.stride, self.padding, self.dilation, self.groups) x = self.relu(x) return x quant_type_to_prepare_expected_node_occurrence = { QuantType.DYNAMIC: {}, # There should be 3 observers: after input, weight and activation. QuantType.STATIC: { ns.call_module(torch.quantization.HistogramObserver): 2, ns.call_module(torch.quantization.PerChannelMinMaxObserver): 1, }, # There should be 3 observers: after input, weight and activation. QuantType.QAT: { ns.call_module(torch.quantization.FakeQuantize): 3, }, } quant_type_to_qconv_fun = { QuantType.STATIC: { 1: ns.call_function(torch.ops.quantized.conv1d), 2: ns.call_function(torch.ops.quantized.conv2d), 3: ns.call_function(torch.ops.quantized.conv3d) }, QuantType.QAT: { 1: ns.call_function(torch.ops.quantized.conv1d), 2: ns.call_function(torch.ops.quantized.conv2d), 3: ns.call_function(torch.ops.quantized.conv3d) }, } quant_type_to_qconv_relu_fun = { QuantType.STATIC: { 1: ns.call_function(torch.ops.quantized.conv1d_relu), 2: ns.call_function(torch.ops.quantized.conv2d_relu), 3: ns.call_function(torch.ops.quantized.conv3d_relu) }, QuantType.QAT: { 1: ns.call_function(torch.ops.quantized.conv1d_relu), 2: ns.call_function(torch.ops.quantized.conv2d_relu), 3: ns.call_function(torch.ops.quantized.conv3d_relu) }, } options = itertools.product( [1, 2, 3], # dims self.static_quant_types, (True, False), # use_bias (True, False), # has_relu (True, False), # functional relu ) for dim, quant_type, use_bias, has_relu, f_relu in options: data_dims = [2, 3] + [4] * dim data = (torch.randn(tuple(data_dims), dtype=torch.float),) model = FuncConv(dim, use_bias, has_relu, f_relu) if has_relu: qconv_fun = quant_type_to_qconv_relu_fun[quant_type][dim] else: qconv_fun = quant_type_to_qconv_fun[quant_type][dim] convert_node_occurrence = { ns.call_function(torch.quantize_per_tensor): 1, qconv_fun: 1, ns.call_method("dequantize"): 1 } prepare_expected_node_occurrence = \ quant_type_to_prepare_expected_node_occurrence[quant_type] self.checkGraphModeFxOp( model, data, quant_type, qconv_fun, prepare_expected_node_occurrence=prepare_expected_node_occurrence, expected_node_occurrence=convert_node_occurrence) @skipIfNoFBGEMM def test_quantized_conv_relu(self): """tests for conv1d_relu/conv2d_relu/conv3d_relu""" conv_module = {1 : torch.nn.Conv1d, 2 : torch.nn.Conv2d, 3 : torch.nn.Conv3d} class ConvNdRelu(torch.nn.Module): def __init__(self, dim, inplace): super(ConvNdRelu, self).__init__() self.conv = conv_module[dim](3, 3, 3).float() self.relu = torch.nn.ReLU(inplace) def forward(self, x): return self.relu(self.conv(x)) class ConvNdFunctionalRelu(torch.nn.Module): def __init__(self, dim): super(ConvNdFunctionalRelu, self).__init__() self.conv = conv_module[dim](3, 3, 3).float() def forward(self, x): return F.relu(self.conv(x)) class ConvNdInplaceFunctionalRelu(torch.nn.Module): def __init__(self, dim): super(ConvNdInplaceFunctionalRelu, self).__init__() self.conv = conv_module[dim](3, 3, 3).float() def forward(self, x): return F.relu(self.conv(x), True) options = itertools.product([1, 2, 3], self.static_quant_types) quantized_nodes = { # dim 1: ns.call_module(nniq.ConvReLU1d), 2: ns.call_module(nniq.ConvReLU2d), 3: ns.call_module(nniq.ConvReLU3d), } for dim, quant_type in options: for m in [ConvNdRelu(dim, True), ConvNdRelu(dim, False), ConvNdFunctionalRelu(dim), ConvNdInplaceFunctionalRelu(dim)]: self.checkGraphModeFxOp( m, self.img_data_dict[dim], quant_type, quantized_nodes[dim]) def _test_quantized_binary_op_impl(self, binary_op, ibinary_op, quantized_op): class Op(torch.nn.Module): def __init__(self, is_inplace, is_scalar): super(Op, self).__init__() self.conv1 = torch.nn.Conv2d(1, 1, 1).float() self.conv2 = torch.nn.Conv2d(1, 1, 1).float() self.is_scalar = is_scalar self.op = ibinary_op if is_inplace else binary_op def forward(self, x, y): x = self.conv1(x) y = 3 if self.is_scalar else self.conv2(y) # x = x + y x = self.op(x, y) # x = y + x x = self.op(y, x) return x # TODO: decide whether we want to quantize or not # in this case # class NonQuantizedOp(torch.nn.Module): # def __init__(self, is_inplace, is_scalar): # super(NonQuantizedOp, self).__init__() # self.is_scalar = is_scalar # self.op = ibinary_op if is_inplace else binary_op # def forward(self, x, y): # y = 3 if self.is_scalar else y # x = self.op(x, y) # return x data = (torch.randn(1, 1, 1, 1, dtype=torch.float), torch.randn(1, 1, 1, 1, dtype=torch.float)) quantized_node = ns.call_function(quantized_op) options = itertools.product([True, False], [True, False]) quant_type = QuantType.STATIC for is_inplace, is_scalar in options: self.checkGraphModeFxOp( Op(is_inplace, is_scalar), data, quant_type, quantized_node) def _test_quantized_binary_op_relu_impl(self, binary_op, ibinary_op, quantized_op): class OpRelu(torch.nn.Module): def __init__(self, is_inplace, is_functional_relu, is_scalar): super(OpRelu, self).__init__() self.conv1 = torch.nn.Conv2d(1, 1, 1).float() self.conv2 = torch.nn.Conv2d(1, 1, 1).float() self.op = ibinary_op if is_inplace else binary_op self.is_functional_relu = is_functional_relu self.is_scalar = is_scalar self.relu = F.relu if self.is_functional_relu \ else torch.nn.ReLU() def forward(self, x, y): x = self.conv1(x) y = 3 if self.is_scalar else self.conv2(y) x = self.op(x, y) x = self.relu(x) x = self.op(y, x) x = self.relu(x) return x data = (torch.rand((1, 1, 1, 1), dtype=torch.float), torch.rand((1, 1, 1, 1), dtype=torch.float)) quant_type = QuantType.STATIC quantized_node = ns.call_function(quantized_op) options = itertools.product( [True, False], [True, False], [True, False]) for is_inplace_op, is_functional_relu, is_scalar in options: self.checkGraphModeFxOp( OpRelu(is_inplace_op, is_functional_relu, is_scalar), data, quant_type, quantized_node) @skipIfNoFBGEMM def test_quantized_add(self): self._test_quantized_binary_op_impl( operator.add, operator.iadd, torch.ops.quantized.add) @skipIfNoFBGEMM def test_quantized_mul(self): self._test_quantized_binary_op_impl( operator.mul, operator.imul, torch.ops.quantized.mul) @skipIfNoFBGEMM def test_quantized_add_relu(self): self._test_quantized_binary_op_relu_impl( operator.add, operator.iadd, torch.ops.quantized.add_relu) @skipIfNoFBGEMM def test_quantized_mul_relu(self): self._test_quantized_binary_op_relu_impl( operator.mul, operator.imul, torch.ops.quantized.mul_relu) # TODO(future PR): make more generic def _test_quantized_add_mul_qat(self, model, expected_node_occurrence): qconfig_dict = {'': torch.quantization.get_default_qat_qconfig('fbgemm')} mp = torch.quantization.quantize_fx.prepare_qat_fx(model, qconfig_dict) self.checkGraphModuleNodes( mp, expected_node_occurrence=expected_node_occurrence) @skipIfNoFBGEMM def test_quantized_add_qat(self): class M(torch.nn.Module): def __init__(self): super().__init__() self.conv1 = torch.nn.Conv2d(1, 1, 1) self.conv2 = torch.nn.Conv2d(1, 1, 1) def forward(self, x): x = torch.add(x, 1.0) x = self.conv1(x) x = torch.add(x, 1.0) x = torch.relu(x) x = self.conv2(x) return x m = M() expected_node_occurrence = { ns.call_module(torch.quantization.FakeQuantize): 4, } self._test_quantized_add_mul_qat(m, expected_node_occurrence) @skipIfNoFBGEMM def test_quantized_mul_qat(self): class M(torch.nn.Module): def __init__(self): super().__init__() self.conv1 = torch.nn.Conv2d(1, 1, 1) self.conv2 = torch.nn.Conv2d(1, 1, 1) def forward(self, x): x = torch.mul(x, 1.0) x = self.conv1(x) x = torch.mul(x, 1.0) x = torch.relu(x) x = self.conv2(x) return x m = M() expected_node_occurrence = { ns.call_module(torch.quantization.FakeQuantize): 4, } self._test_quantized_add_mul_qat(m, expected_node_occurrence) def test_int8_input_no_unnecessary_fq(self): """ If the inputs to the graph are quantized and the only node does not need an activation observer, verifies that the activation observer is not inserted. """ class M(nn.Module): def __init__(self, scalar): super().__init__() self.scalar = scalar self.add_func = torch.nn.quantized.FloatFunctional() def forward(self, x): return self.add_func.add_scalar(x, self.scalar) m = M(0.5) mp = torch.quantization.quantize_fx.prepare_qat_fx( m, {'': torch.quantization.get_default_qat_qconfig('fbgemm')}, prepare_custom_config_dict={"input_quantized_idxs": [0]}) expected_node_occurrence = { ns.call_module(torch.quantization.FakeQuantize): 0, } self.checkGraphModuleNodes( mp, expected_node_occurrence=expected_node_occurrence) def test_quant_output_always_observed(self): """ If the output is hardcoded to be quantized, ensure that there is always an observer, even if the last non-output node is not quantizeable. """ qconfig_dict = {'': torch.quantization.get_default_qat_qconfig('fbgemm')} prepare_custom_config_dict = {'output_quantized_idxs': [0]} data = (torch.randn(4, 1, 4, 4),) # non-quantizeable node, quantized output class M1(torch.nn.Module): def __init__(self): super().__init__() self.identity = torch.nn.Identity() def forward(self, x): x = self.identity(x) return x m1 = M1() self.checkGraphModeFxOp( m1, data, QuantType.QAT, prepare_expected_node_occurrence={ ns.call_module(torch.quantization.FakeQuantize): 1, }, expected_node_occurrence={ ns.call_function(torch.quantize_per_tensor): 1, }, prepare_custom_config_dict=prepare_custom_config_dict) # quantizeable node, quantized output class M2(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(1, 1, 1) def forward(self, x): x = self.conv(x) return x m2 = M2() self.checkGraphModeFxOp( m2, data, QuantType.QAT, prepare_expected_node_occurrence={ # one for weights, one for activations ns.call_module(torch.quantization.FakeQuantize): 2, }, expected_node_occurrence={ ns.call_function(torch.quantize_per_tensor): 1, }, prepare_custom_config_dict=prepare_custom_config_dict) @skipIfNoFBGEMM def test_quantized_cat(self): """ quantization of the output of cat will be depend on the input of cat. we only quantize the output of cat when its inputs are quantized. """ class QuantizedCat(torch.nn.Module): def __init__(self): super(QuantizedCat, self).__init__() self.conv1 = torch.nn.Conv2d(2, 2, 2).float() self.conv2 = torch.nn.Conv2d(2, 2, 2).float() def forward(self, x, y): x = self.conv1(x) y = self.conv2(y) return torch.cat([x, y], 1) # TODO: decide whether to quantize in this case # class NonQuantizedCat(torch.nn.Module): # def __init__(self): # super(NonQuantizedCat, self).__init__() # def forward(self, x, y): # return torch.cat([x, y], 1) data = (torch.randn(1, 2, 5, 5, dtype=torch.float), torch.randn(1, 2, 5, 5, dtype=torch.float)) quantized_node = ns.call_function(torch.ops.quantized.cat) for quant_type in self.static_quant_types: self.checkGraphModeFxOp(QuantizedCat(), data, quant_type, quantized_node) @skipIfNoFBGEMM def test_qbatch_norm(self): bn_module = { # TODO: quantized batchnorm 1d module is missing # 1 : torch.nn.BatchNorm1d, 2 : torch.nn.BatchNorm2d, 3 : torch.nn.BatchNorm3d, } class M(torch.nn.Module): def __init__(self, dim): super(M, self).__init__() self.bn = bn_module[dim](3).to(torch.float) def forward(self, x): return self.bn(x) options = itertools.product(self.static_quant_types, [2, 3]) quantized_nodes = { # 1: ns.call_module(nnq.BatchNorm1d), 2: ns.call_module(nnq.BatchNorm2d), 3: ns.call_module(nnq.BatchNorm3d), } for quant_type, dim in options: model = self.checkGraphModeFxOp( M(dim), self.img_data_dict[dim], quant_type, quantized_nodes[dim]) @skipIfNoFBGEMM def test_qbatch_norm_relu(self): bn_module = {2 : torch.nn.BatchNorm2d, 3 : torch.nn.BatchNorm3d} class BNRelu(torch.nn.Module): def __init__(self, dim, inplace): super(BNRelu, self).__init__() self.bn = bn_module[dim](3).to(torch.float) self.relu = torch.nn.ReLU(inplace=inplace) def forward(self, x): return self.relu(self.bn(x)) class BNFuncRelu(torch.nn.Module): def __init__(self, dim): super(BNFuncRelu, self).__init__() self.bn = bn_module[dim](3).to(torch.float) def forward(self, x): return F.relu(self.bn(x), False) class BNFuncInplaceRelu(torch.nn.Module): def __init__(self, dim): super(BNFuncInplaceRelu, self).__init__() self.bn = bn_module[dim](3).to(torch.float) def forward(self, x): return F.relu(self.bn(x), True) options = itertools.product(self.static_quant_types, [2, 3]) quantized_nodes = { 2: ns.call_module(nniq.BNReLU2d), 3: ns.call_module(nniq.BNReLU3d), } for quant_type, dim in options: for instance in [BNRelu(dim, True), BNRelu(dim, False), BNFuncRelu(dim), BNFuncInplaceRelu(dim)]: self.checkGraphModeFxOp( instance, self.img_data_dict[dim], quant_type, quantized_nodes[dim]) def _test_activation_impl( self, float_module, float_op, quantized_module, quantized_op): ''' Test for activation op(with inplace options), float_op can be torch op or functional op ''' class M(torch.nn.Module): def __init__(self, is_module, inplace): super(M, self).__init__() self.is_module = is_module self.inplace = inplace if self.is_module: self.op = float_module(self.inplace) else: self.op = float_op def forward(self, input): if self.is_module: return self.op(input) else: return self.op(input, self.inplace) options = itertools.product([True, False], [True, False], self.static_quant_types) quantized_nodes = { # is_module True: ns.call_module(quantized_module), False: ns.call_function(quantized_op), } for is_module, is_inplace, quant_type in options: self.checkGraphModeFxOp( M(is_module, is_inplace), self.img_data_2d, quant_type, quantized_nodes[is_module]) def test_hardswish(self): self._test_activation_impl(nn.Hardswish, F.hardswish, nnq.Hardswish, torch.ops.quantized.hardswish) def test_elu(self): self._test_activation_impl(nn.ELU, F.elu, nnq.ELU, torch.ops.quantized.elu) def test_leaky_relu(self): self._test_activation_impl(nn.LeakyReLU, F.leaky_relu, nnq.LeakyReLU, torch.ops.quantized.leaky_relu) def _test_norm_impl( self, float_module, float_op, op_args, data, quantized_module, quantized_op, skip_op_arg_for_functional=False): ''' Test for normalization op, float_op can be torch op or functional op, op_args is a list of positional argument for the module/op ''' class M(torch.nn.Module): def __init__(self, is_module): super(M, self).__init__() self.is_module = is_module if self.is_module: self.op = float_module(*op_args) else: self.op = float_op def forward(self, input): if self.is_module: return self.op(input) else: args = [input] if not skip_op_arg_for_functional: args += op_args return self.op(*args) options = itertools.product([True, False], self.static_quant_types) quantized_nodes = { # is_module True: ns.call_module(quantized_module), False: ns.call_function(quantized_op), } for is_module, quant_type in options: self.checkGraphModeFxOp( M(is_module), data, quant_type, quantized_nodes[is_module]) def test_layer_norm(self): data = (torch.rand((1, 2, 5, 5), dtype=torch.float),) self._test_norm_impl( nn.LayerNorm, F.layer_norm, [[2, 5, 5]], data, nnq.LayerNorm, torch.ops.quantized.layer_norm) def test_instance_norm(self): data_1d = (torch.rand((1, 4, 5), dtype=torch.float),) data_2d = (torch.rand((1, 4, 5, 1), dtype=torch.float),) data_3d = (torch.rand((1, 4, 5, 1, 1), dtype=torch.float),) data_dict = {1 : data_1d, 2 : data_2d, 3 : data_3d} instance_norm_modules = {1 : nn.InstanceNorm1d, 2 : nn.InstanceNorm2d, 3 : nn.InstanceNorm3d} quantized_instance_norm_modules = { 1 : nnq.InstanceNorm1d, 2 : nnq.InstanceNorm2d, 3 : nnq.InstanceNorm3d } for dim in [1, 2, 3]: data = data_dict[dim] module = instance_norm_modules[dim] quantized_module = quantized_instance_norm_modules[dim] self._test_norm_impl( module, F.instance_norm, [4], data, quantized_module, torch.ops.quantized.instance_norm, skip_op_arg_for_functional=True) @skipIfNoFBGEMM def test_clamp(self): class M(torch.nn.Module): def __init__(self): super(M, self).__init__() self.conv = torch.nn.Conv2d(2, 2, 2).float() self.relu6 = torch.nn.ReLU6() self.relu6_ = torch.nn.ReLU6(True) self.hardtanh = torch.nn.Hardtanh() self.hardtanh_ = torch.nn.Hardtanh(inplace=True) def forward(self, x): x = self.conv(x) x = self.relu6(x) self.relu6_(x) x = F.relu6(x) x = torch.clamp(x, -3, 3) x = x.clamp(-2.5, 2.5) # x = x.clamp_(-2, 2) # Enable when quantized `clamp_` is ready x = self.hardtanh(x) self.hardtanh_(x) x = F.hardtanh(x) F.hardtanh_(x) return x data = (torch.rand((1, 2, 5, 5), dtype=torch.float),) # list of node that should occur in order node_list = [ ns.call_function(torch.quantize_per_tensor), ns.call_module(nnq.Conv2d), ns.call_function(F.hardtanh_), ns.call_method('dequantize') ] for quant_type in self.static_quant_types: m = self.checkGraphModeFxOp( M(), data, quant_type, expected_node_list=node_list) @skipIfNoFBGEMM def test_general_shape_ops(self): """ A test that checks dequantize will be swapped for all supported general shape ops like aten::flatten without actually checking for execution of these ops """ class M(torch.nn.Module): def __init__(self): super(M, self).__init__() self.maxpool1d = torch.nn.MaxPool1d(kernel_size=3) self.maxpool2d = torch.nn.MaxPool2d(kernel_size=3) self.maxpool3d = torch.nn.MaxPool3d(kernel_size=3) self.dropout = torch.nn.Dropout() self.conv1 = torch.nn.Conv2d(3, 3, 3) self.conv2 = torch.nn.Conv2d(3, 3, 3) self.relu = torch.nn.ReLU() def forward(self, x): x = self.conv1(x) # add_scalar x = x + 3 # mul_scalar x = x * 3 # add_scalar_out x += 3 # mul_scalar_out x *= 3 # add_scalar_relu x = x + 3 x = F.relu(x) # add_scalar_relu_out x += 3 x = F.relu(x) # mul_scalar_relu x = x * 3 x = F.relu(x) # mul_scalar_relu_out x *= 3 x = F.relu(x) x = self.maxpool1d(x) x = self.maxpool2d(x) x = self.maxpool3d(x) x = torch.flatten(x) x = torch.max(x) x = torch.min(x) x = x.reshape([-1]) x = x.resize_(1, 1, x.numel()) x = x.view(-1) # prim::ListConstruct xs = [x, x] # prim::ListUnpack x, y = xs # prim::TupleConstruct xs = (x, x) # prim::TupleUnpack x, y = xs x = x.transpose(1, 2) x = x.contiguous() x, y = torch.chunk(x, 2) x = F.dropout(x) x = self.dropout(x) x, _ = torch.sort(x) x = x.permute(0, 2, 3, 1) x = x.repeat_interleave(3, 1) x = torch.repeat_interleave(x, 3, 1) x = self.relu(x) x = F.relu(x) x = F.relu(x, inplace=True) x = x.relu() x.relu_() x = x.squeeze(0) x.squeeze_(0) x = torch.squeeze(x, 0) x = x.unsqueeze(0) x.unsqueeze_(0) x = torch.unsqueeze(x, 0) x = x.detach() x.detach_() x = x.repeat(4, 2) y = [] y.append(x) z = torch.stack(y, 0) z = [z, z] x, _ = z x = self.conv2(x) return x data = torch.rand(1, 3, 10, 10) # This model is not executable since we just put all ops # in the same forward m = M().eval() # nothing to fuse so skipping the fuse step qconfig_dict = {'': default_qconfig} prepared = prepare_fx(m, qconfig_dict) # not runnable quantized = convert_fx(prepared) # This checks that the dequantize from the output of first conv # is being propagated to the end, so that we don't insert extra # observers and also successfully fused two quantized::conv2d # patterns # one quantize_per_tensor for input # check exact counts of quantize and dequantize count_check = { ns.call_function(torch.quantize_per_tensor) : 1, ns.call_method('dequantize') : 1 } order_check = [ ns.call_function(torch.quantize_per_tensor), ns.call_module(nnq.Conv2d), ns.call_module(nnq.Conv2d), ns.call_method('dequantize'), ] self.checkGraphModuleNodes( quantized, expected_node_occurrence=count_check, expected_node_list=order_check) @skipIfNoFBGEMM def test_general_value_ops(self): """ A test that checks correct patterns are produced for all supported general value ops like aten::avg_pool2d \ without actually checking for execution of these ops """ class M(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(3, 3, 3) self.avg_pool1d = torch.nn.AvgPool1d(3) self.avg_pool2d = torch.nn.AvgPool2d(3) self.avg_pool3d = torch.nn.AvgPool3d(3) self.adaptive_avg_pool1d = torch.nn.AdaptiveAvgPool1d((1)) self.adaptive_avg_pool2d = torch.nn.AdaptiveAvgPool2d((1, 1)) self.adaptive_avg_pool3d = torch.nn.AdaptiveAvgPool3d((1, 1, 1)) def forward(self, x): x = self.conv(x) x = self.avg_pool1d(x) x = self.avg_pool2d(x) x = self.avg_pool3d(x) x = self.adaptive_avg_pool1d(x) x = self.adaptive_avg_pool2d(x) x = self.adaptive_avg_pool3d(x) x = F.avg_pool1d(x, 3) x = F.avg_pool2d(x, 3) x = F.avg_pool3d(x, 3) x = F.adaptive_avg_pool1d(x, (1)) x = F.adaptive_avg_pool2d(x, (1, 1)) x = F.adaptive_avg_pool3d(x, (1, 1, 1)) x = torch.mean(x) x = torch.mean(x, [2, 3], False) x = x.mean() x = x.mean([2, 3], True) x = F.interpolate(x, 4, mode='nearest') x = F.interpolate(x, 4, mode='linear') x = self.conv(x) return x # This model is not executable since we just put all ops # in the same forward m = M().eval() # nothing to fuse so skipping the fuse step qconfig_dict = {'': default_qconfig} prepared = prepare_fx(m, qconfig_dict) # not runnable quantized = convert_fx(prepared) # This checks that the dequantize from the output of first conv # is being propagated to the end, so that we don't insert extra # observers # check exact counts of quantize and dequantize count_check = { ns.call_function(torch.quantize_per_tensor) : 1, ns.call_method('dequantize') : 1 } order_check = [ ns.call_function(torch.quantize_per_tensor), ns.call_module(nnq.Conv2d), ns.call_module(nnq.Conv2d), ns.call_method('dequantize'), ] self.checkGraphModuleNodes( quantized, expected_node_occurrence=count_check, expected_node_list=order_check) @skipIfNoFBGEMM def test_fixed_qparams_ops(self): class M(torch.nn.Module): def __init__(self): super().__init__() self.conv = torch.nn.Conv2d(3, 3, 3) self.sigmoid = torch.nn.Sigmoid() self.hardsigmoid = torch.nn.Hardsigmoid() self.tanh = torch.nn.Tanh() def forward(self, x): x = self.conv(x) # F.sigmoid is deprecated x = self.sigmoid(x) x = torch.sigmoid(x) x = x.sigmoid() x.sigmoid_() x = self.hardsigmoid(x) x = F.hardsigmoid(x) x = F.hardsigmoid(x, inplace=True) x = x.hardsigmoid() x.hardsigmoid_() x = self.tanh(x) # F.tanh is deprecated x = torch.tanh(x) x = x.tanh() x.tanh_() x = self.conv(x) return x for eval_mode in [True, False]: # This model is not executable since we just put all ops # in the same forward m = M() if eval_mode: m.eval() qconfig = default_qconfig prepare = prepare_fx fq_count = 0 else: m.train() qconfig = default_qat_qconfig prepare = prepare_qat_fx fq_count = 13 # nothing to fuse so skipping the fuse step qconfig_dict = {'': qconfig} prepared = prepare(m, qconfig_dict) # check the correct number of activation_post_process is inserted count_check = { ns.call_module(FixedQParamsFakeQuantize) : fq_count, } self.checkGraphModuleNodes( prepared, expected_node_occurrence=count_check) # not runnable quantized = convert_fx(prepared) # This checks that the dequantize from the output of first conv # is being propagated to the end, so that we don't insert extra # observers # check exact counts of quantize and dequantize count_check = { ns.call_function(torch.quantize_per_tensor) : 1, ns.call_method('dequantize') : 1 } order_check = [ ns.call_function(torch.quantize_per_tensor), ns.call_module(nnq.Conv2d), ns.call_module(nn.Sigmoid), ns.call_module(nnq.Conv2d), ns.call_method('dequantize'), ] self.checkGraphModuleNodes( quantized, expected_node_occurrence=count_check, expected_node_list=order_check) def test_float_functional(self): class TorchAdd(nn.Module): """Wrapper around torch.add so that all ops can be found at build""" def __init__(self): super().__init__() self.add_func = nnq.FloatFunctional() def forward(self, x, y): return self.add_func.add(x, y) class M(torch.nn.Module): def __init__(self): super().__init__() self.ff1 = TorchAdd() self.ff2 = nnq.FloatFunctional() self.ff3 = nnq.FloatFunctional() self.ff4 = nnq.FloatFunctional() self.ff5 = nnq.FloatFunctional() self.ff6 = nnq.FloatFunctional() def forward(self, x): x = self.ff1(x, x) x = self.ff2.add_scalar(x, 3) x = self.ff3.mul(x, x) x = self.ff4.mul_scalar(x, 3) x = self.ff5.add_relu(x, x) x = self.ff6.cat([x]) return x data = torch.rand(3, 3) # Note: QAT test succeeded by chance, to make it actually work # we need to fix eager mode FloatFunctional by removing # activation_post_process in add_scalar and mul_scalar for quant_type in self.static_quant_types: m = M() ref_m = torch.quantization.QuantWrapper(M()) is_qat = quant_type == QuantType.QAT if is_qat: m.train() ref_m.train() qconfig = default_qat_qconfig expected_act_post_process = torch.quantization.FakeQuantize else: m.eval() ref_m.eval() qconfig = default_qconfig expected_act_post_process = torch.quantization.MinMaxObserver prepare_fx_function = prepare_qat_fx if is_qat else prepare_fx qconfig_dict = {"": qconfig} m = prepare_fx_function(m, qconfig_dict) node_occurrence = { ns.call_module(expected_act_post_process): 5, ns.call_module(torch.nn.quantized.FloatFunctional): 0 } self.checkGraphModuleNodes(m, expected_node_occurrence=node_occurrence) m(data) node_list = [ ns.call_function(torch.quantize_per_tensor), ns.call_function(torch.ops.quantized.add), ns.call_function(torch.ops.quantized.add), ns.call_function(torch.ops.quantized.mul), ns.call_function(torch.ops.quantized.mul), ns.call_function(torch.ops.quantized.add_relu), ns.call_function(torch.ops.quantized.cat), ns.call_method('dequantize') ] m = convert_fx(m) self.checkGraphModuleNodes(m, expected_node_list=node_list) # make sure numerics match with eager mode ref_m.qconfig = qconfig prepare_function = prepare_qat if is_qat else prepare ref_m = prepare_function(ref_m) ref_m(data) ref_m = convert(ref_m) self.assertEqual(m(data), ref_m(data)) def test_embedding(self): class M(torch.nn.Module): def __init__(self): super().__init__() self.emb = torch.nn.Embedding(num_embeddings=10, embedding_dim=12) def forward(self, indices): return self.emb(indices) model = M().eval() indices = torch.tensor([9, 6, 5, 7, 8, 8, 9, 2, 8, 6, 6, 9, 1, 6, 8, 8, 3, 2, 3, 6, 3, 6, 5, 7, 0, 8, 4, 6, 5, 8, 2, 3]) quantized_node = ns.call_module(nnq.Embedding) configs = [ (float_qparams_weight_only_qconfig, ns.call_module(nnq.Embedding)), (None, ns.call_module(nn.Embedding)), (default_qconfig, ns.call_module(nn.Embedding)), ] for qconfig, node in configs: qconfig_dict = {"": qconfig} m = prepare_fx(model, qconfig_dict) self.checkGraphModuleNodes(m, expected_node_occurrence={ ns.call_module(torch.quantization.MinMaxObserver): 0 }) m = convert_fx(m) self.checkGraphModuleNodes(m, expected_node=node) # make sure it runs m(indices) def test_embedding_bag(self): class M(torch.nn.Module): def __init__(self): super().__init__() self.emb = torch.nn.EmbeddingBag(num_embeddings=10, embedding_dim=12, include_last_offset=True) def forward(self, indices, offsets): return self.emb(indices, offsets) indices = torch.tensor([9, 6, 5, 7, 8, 8, 9, 2, 8, 6, 6, 9, 1, 6, 8, 8, 3, 2, 3, 6, 3, 6, 5, 7, 0, 8, 4, 6, 5, 8, 2, 3]) offsets = torch.tensor([0, 19, 20, 28, 28, 32]) quantized_node = ns.call_module(nnq.EmbeddingBag) inputs = (indices, offsets) for dtype in [torch.quint8, torch.quint4x2]: model = M().eval() float_qparams_observer = PerChannelMinMaxObserver.with_args(dtype=dtype, qscheme=torch.per_channel_affine_float_qparams, ch_axis=0) float_qparams_qconfig = QConfigDynamic(activation=default_placeholder_observer, weight=float_qparams_observer) self.checkGraphModeFxOp( model, inputs, QuantType.DYNAMIC, quantized_node, custom_qconfig=float_qparams_qconfig ) # check it works in None and static qconfig for qconfig in [None, default_qconfig]: qconfig_dict = {"": default_qconfig} m = M().eval() m = prepare_fx(model, qconfig_dict) self.checkGraphModuleNodes(m, expected_node_occurrence={ ns.call_module(torch.quantization.MinMaxObserver): 0 }) m = convert_fx(m) self.checkGraphModuleNodes(m, expected_node=ns.call_module(nn.EmbeddingBag)) # make sure it runs m(*inputs) def _test_rnn_impl(self, qconfigs, M, module_type_strs, module_types, sample_input): options = itertools.product(qconfigs, module_type_strs) for qconfig, module_type_str in options: model_eager = M(module_type_str).eval() model_graph = copy.deepcopy(model_eager) if torch.backends.quantized.engine == 'qnnpack' and \ qconfig is float16_dynamic_qconfig: continue # fp16 dynamic quant is not supported for qnnpack eager_qconfig_dict = {x : qconfig for x in module_types} model_eager = quantize_dynamic(model_eager, qconfig_spec=eager_qconfig_dict) graph_qconfig_dict = { "object_type": [ (x, qconfig) for x in module_types ] } model_graph = prepare_fx(model_graph, graph_qconfig_dict) model_graph = convert_fx(model_graph) self.assertEqual(model_eager(sample_input), model_graph(sample_input)) self.checkScriptable(model_graph, [[sample_input]], True) def test_rnn_cell(self): qconfigs = [per_channel_dynamic_qconfig, default_dynamic_qconfig, float16_dynamic_qconfig] module_type_strs = ['LSTMCell', 'GRUCell', 'RNNTanh', 'RNNReLU'] module_types = [torch.nn.LSTMCell, torch.nn.GRUCell, torch.nn.RNNCell] sample_input = torch.tensor([[100, -155], [-155, 100], [100, -155]], dtype=torch.float) self._test_rnn_impl(qconfigs, RNNCellDynamicModel, module_type_strs, module_types, sample_input) def test_rnn(self): qconfigs = [per_channel_dynamic_qconfig, default_dynamic_qconfig, float16_dynamic_qconfig] module_type_strs = ['LSTM'] module_types = [torch.nn.LSTM] niter = 10 sample_input = torch.tensor([[100, -155], [-155, 100], [100, -155]], dtype=torch.float).unsqueeze(0).repeat(niter, 1, 1) self._test_rnn_impl(qconfigs, RNNDynamicModel, module_type_strs, module_types, sample_input) def _test_conv_transpose_impl( self, float_cls: Callable, q_cls: Callable, data: torch.Tensor): with override_quantized_engine('qnnpack'): # Create fp32 versions of FX and Eager models m1 = torch.nn.Sequential(float_cls(1, 1, 1)) m2 = torch.nn.Sequential(float_cls(1, 1, 1)) m2.load_state_dict(m1.state_dict()) m2 = torch.quantization.QuantWrapper(m2) # FX graph q_result1 = self.checkGraphModeFxOp( m1, (data,), QuantType.STATIC, expected_node_occurrence={ ns.call_module(q_cls): 1, }) # Eager m2.qconfig = get_default_qconfig(torch.backends.quantized.engine) m2.eval() m2p = torch.quantization.prepare(m2) m2p(data) m2q = torch.quantization.convert(m2p) q_result2 = m2q(data) # verify results match self.assertTrue(torch.allclose(q_result1, q_result2)) @unittest.skipUnless('qnnpack' in supported_qengines, "This Pytorch Build has not been built with or does not support QNNPACK") def test_conv_transpose_1d(self): self._test_conv_transpose_impl( torch.nn.ConvTranspose1d, nnq.ConvTranspose1d, torch.randn(4, 1, 4)) @unittest.skipUnless('qnnpack' in supported_qengines, "This Pytorch Build has not been built with or does not support QNNPACK") def test_conv_transpose_2d(self): self._test_conv_transpose_impl( torch.nn.ConvTranspose2d, nnq.ConvTranspose2d, torch.randn(4, 1, 4, 4)) class TestQuantizeFxModels(QuantizationTestCase): def _test_model_impl( self, mode, name, model, eager_quantizable_model, check_with_eager=True, diff_of_quant=None, diff_from_eager=None): if diff_of_quant is None or diff_from_eager is None: diff_of_quant = {} diff_from_eager = {} if mode not in diff_of_quant or mode not in diff_from_eager: diff_of_quant[mode] = {} diff_from_eager[mode] = {} input_tensor = torch.rand(1, 3, 224, 224) input_tensor_inception = torch.rand(1, 3, 299, 299) output_value = torch.randint(0, 1, (1,)) # print('quantizing:', name, ' mode:', mode) if name == 'inception_v3': input_value = input_tensor_inception else: input_value = input_tensor qconfig = default_qconfig if mode == 'static' else default_qat_qconfig qconfig_dict = {'': qconfig} # print('graph module:', graph_module.src) script = torch.jit.script(model) # make sure graph module and script module are both runanble original_out = model(input_value) is_not_tuple_out = not isinstance(original_out, tuple) script_out = script(input_value) # set to train just before quantization prepare_fx_fn = prepare_fx if mode != 'static': model.train() prepare_fx_fn = prepare_qat_fx prepared = prepare_fx_fn(model, qconfig_dict) if mode == 'ddp': mp.spawn(run_ddp, args=(world_size, prepared), nprocs=world_size, join=True) elif mode == 'qat': assert prepared.training, 'prepared must be in training mode for qat' optimizer = torch.optim.SGD(prepared.parameters(), lr=0.0001) criterion = nn.CrossEntropyLoss() train_one_epoch(prepared, criterion, optimizer, [(input_value, output_value)], torch.device('cpu'), 1) else: for i in range(10): prepared(input_value) # print('after observation root:', prepared.root) qgraph = convert_fx(prepared) # print('after quantization root:', qgraph.root) # print('after quantization code:', qgraph.src) qgraph.eval() qgraph_script = torch.jit.script(qgraph) # print('quantized and scripted:', qgraph_script.graph) qgraph_out = qgraph(input_value) qgraph_script = qgraph_script(input_value) if is_not_tuple_out: diff_of_quant[mode][name] = (original_out - qgraph_out).abs().max() assert torch.allclose(qgraph_out, qgraph_script), 'graph, scripted graph' else: print('tuple output') if eager_quantizable_model is not None: # comparing to eager mode quantization qeager = eager_quantizable_model ref_out = qeager(input_value) qeager.qconfig = qconfig if mode == 'static': qeager.fuse_model() prepare(qeager, inplace=True) else: qeager.train() qeager.fuse_model() prepare_qat(qeager, inplace=True) # calibration if mode == 'ddp': mp.spawn(run_ddp, args=(world_size, qeager), nprocs=world_size, join=True) elif mode == 'qat': assert qeager.training, 'qeager should be in training mode for qat' optimizer = torch.optim.SGD(qeager.parameters(), lr=0.0001) train_one_epoch(qeager, criterion, optimizer, [(input_value, output_value)], torch.device('cpu'), 1) else: for i in range(10): qeager(input_value) # print('ref after observation:', qeager) convert(qeager, inplace=True) qeager.eval() # print('ref after quantization:', qeager) qeager_out = qeager(input_value) qeager_script = torch.jit.script(qeager) qscript_out = qeager_script(input_value) if is_not_tuple_out: diff_from_eager[mode][name] = (qeager_out - qgraph_out).abs().max() if check_with_eager: self.assertEqual(diff_from_eager[mode][name], 0, 'Result of graph mode quantization and ' + 'eager mode quantization on model: ' + name + ' should match. Mode: ' + mode + ' diff:' + str(diff_from_eager[mode][name])) def _test_building_block(self, quant_type, BB): eager = BB().float() graph = copy.deepcopy(eager) if quant_type == QuantType.STATIC: qconfig = default_qconfig eager_prepare = prepare graph_prepare = prepare_fx eager.eval() graph.eval() calibrate_or_train = test_only_eval_fn data = self.img_data_2d else: assert quant_type == QuantType.QAT qconfig = default_qat_qconfig eager_prepare = prepare_qat graph_prepare = prepare_qat_fx eager.train() graph.train() calibrate_or_train = test_only_train_fn data = self.img_data_2d_train if hasattr(eager, "fuse_model"): eager.fuse_model() eager = QuantWrapper(eager) eager.qconfig = qconfig eager = eager_prepare(eager) qconfig_dict = {"": qconfig} graph = graph_prepare(graph, qconfig_dict) eager_out = eager(data[0][0]) graph_out = graph(data[0][0]) self.assertEqual(eager_out, graph_out) calibrate_or_train(eager, data) calibrate_or_train(graph, data) eager = convert(eager) graph = convert_fx(graph) eager_out = eager(data[0][0]) graph_out = graph(data[0][0]) self.assertEqual(eager_out, graph_out) @override_qengines def test_resnet_base(self): models = [ResNetBase] options = itertools.product(self.static_quant_types, models) for quant_type, M in options: self._test_building_block(quant_type, M) @skip_if_no_torchvision @skipIfNoFBGEMM @unittest.skip("skip for now since tbb failed") def test_torchvision(self): from torchvision import models from torchvision.models import quantization as quantized_models def get_available_classification_models(models): return [k for k, v in models.__dict__.items() if callable(v) and k[0].lower() == k[0] and k[0] != "_"] model_list = get_available_classification_models(models) quantized_model_list = get_available_classification_models(quantized_models) no_pretrained_model = set(['shufflenet_v2_x0_5', 'shufflenet_v2_x1_5', 'shufflenet_v2_x2_0']) quantized_model_list = set(quantized_model_list) - no_pretrained_model # test eager and graph consistency model_list = quantized_model_list # inception_v3 is not symbolically traceable: https://github.com/pytorch/pytorch/issues/48813 model_list = set(model_list) - {'inception_v3'} # mobilenet: dropout error RuntimeError: "bernoulli_scalar_cpu_" not implemented for 'QUInt8' # incpetion_v3: looks like there is some problem with AuxLogits quantized_not_working = [('qat', 'inception_v3'), ('static', 'inception_v3')] fx_eager_not_matching = ['googlenet', # because _transform_input is not quantized in eager 'mobilenet_v2'] # because relu6 is replaced as relu in mobilenetv2 diff_of_quant = {} diff_from_eager = {} modes = ['static', 'qat'] options = itertools.product(modes, model_list) for mode, name in options: pretrained = name in quantized_model_list # load pretrained model to compare with quantized model if name in quantized_model_list: if (mode, name) in quantized_not_working: eager_quantizable_model = None else: eager_quantizable_model = quantized_models.__dict__[name](pretrained=True, quantize=False).eval().float() # compare with eager mode quantized model when it is available pretrained = eager_quantizable_model is not None model = models.__dict__[name](pretrained=pretrained).eval().float() check_with_eager = name not in fx_eager_not_matching self._test_model_impl( mode, name, model, eager_quantizable_model, check_with_eager, diff_of_quant, diff_from_eager) def print_diffs(diffs): for mode, diffs_for_mode in diffs.items(): print('mode:', mode) for name, diff in diffs_for_mode.items(): print(name, ':', diff) # print('differences between float and quantized') # print_diffs(diff_of_quant) # print('----------------------') # print('differences between graph mode and eager mode') # print_diffs(diff_from_eager) # print('----------------------') @skip_if_no_torchvision @skip_if_not_multigpu @skipIfNoFBGEMM def test_resnet18_ddp(self): from torchvision import models from torchvision.models import quantization as quantized_models eager_quantizable_model = quantized_models.__dict__[name](pretrained=True, quantize=False).eval().float() model = models.__dict__[name](pretrained=True).eval().float() self._test_model_impl( 'ddp', 'resnet18', model, eager_quantizable_model)