feat(imperative): add TracedModule checker

GitOrigin-RevId: 12de7b278e

imperative/python/megengine/traced_module/__init__.py

@@ -10,6 +10,7 @@ from ..core._imperative_rt.core2 import set_cpp_apply_module_trace
 from . import compat
 from ._passes import optimize
 from .pytree import register_supported_type
+from .tm_config import disable_default_checker, enable_expr_checker
 from .traced_module import (
     TracedModule,
     _register_all_builtin_module,
@@ -29,4 +30,6 @@ __all__ = [
     "wrap",
     "TracedModule",
     "optimize",
+    "enable_expr_checker",
+    "disable_default_checker",
 ]

imperative/python/megengine/traced_module/checker.py

@@ -0,0 +1,142 @@
+# MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
+#
+# Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+import traceback
+from typing import Sequence
+
+import numpy as np
+
+from ..core._imperative_rt.core2 import apply
+from ..core._imperative_rt.ops import ROIAlign, ROIPooling
+from ..core.ops.builtin import Copy
+from ..core.tensor.utils import isscalar, setscalar
+from ..tensor import Tensor
+from .tm_config import _exclude_from_trace
+
+
+class TracedModuleChecker:
+    def __init__(self, tracer):
+        self._active_node2values = []
+        self.tracer = tracer
+
+        self.node_without_tensor_info = {}
+
+    def push_scope(self):
+        self._active_node2values.append({})
+
+    def pop_scope(self):
+        self._active_node2values.pop()
+
+    def current_node2values(self):
+        return self._active_node2values[-1]
+
+    def reset_checker(self):
+        self._active_node2values = []
+
+    def check_node_not_in_scope(self):
+        if self.node_without_tensor_info:
+            for node, info in self.node_without_tensor_info.items():
+                for expr in info[0]._exprs:
+                    if node in expr.inputs or node in expr.outputs:
+                        traceback.print_list(info[1])
+                        raise ValueError(
+                            "node({}) not in the graph:\n{}".format(node, info[0])
+                        )
+            return True
+        else:
+            return False
+
+    def check_net_outputs(self, tm_res, gt_res):
+        if isinstance(tm_res, Tensor):
+            np.testing.assert_allclose(tm_res.numpy(), gt_res.numpy())
+        elif isinstance(tm_res, Sequence):
+            for i, j in zip(tm_res, gt_res):
+                np.testing.assert_allclose(i.numpy(), j.numpy())
+        else:
+            for k in tm_res.__dict__.keys():
+                np.testing.assert_allclose(
+                    getattr(tm_res, k).numpy(), getattr(gt_res, k).numpy()
+                )
+
+    def record_nodemixin(self, node, value):
+        self.current_node2values()[node] = value
+
+    def record_node2value(self, node, value):
+        with _exclude_from_trace():
+            self.current_node2values()[node] = apply(
+                Copy(comp_node=value.device), value
+            )[0]
+            if isscalar(value):
+                setscalar(self.current_node2values()[node])
+
+    def check_apply_special_cases(self, opdef, num_outputs):
+        indexs = list(range(num_outputs))
+        if isinstance(opdef, ROIAlign) and opdef.mode == ROIAlign.Mode.AVERAGE:
+            indexs.pop(-1)
+        if isinstance(opdef, ROIPooling) and opdef.mode == ROIPooling.Mode.AVERAGE:
+            indexs.pop(-1)
+        return indexs
+
+    def check_expr_results(self, expr_outputs, gt_outputs, indexs=None):
+        expr_outputs = (
+            (expr_outputs,) if not isinstance(expr_outputs, Sequence) else expr_outputs
+        )
+        gt_outputs = (
+            (gt_outputs,) if not isinstance(gt_outputs, Sequence) else gt_outputs
+        )
+        if indexs is not None:
+            for i in indexs:
+                np.testing.assert_allclose(
+                    expr_outputs[i].numpy(), gt_outputs[i].numpy()
+                )
+        else:
+            np.testing.assert_allclose(expr_outputs, gt_outputs)
+
+    def get_node2value(self, inputs, start_idx=0):
+        inp_values = []
+        has_node_not_in_scope = False
+        for i in range(start_idx, len(inputs)):
+            try:
+                inp_values.append(self.current_node2values()[inputs[i]])
+            except:
+                has_node_not_in_scope = True
+                self.node_without_tensor_info[inputs[i]] = [
+                    self.tracer.current_scope(),
+                    traceback.extract_stack(),
+                ]
+        return inp_values, has_node_not_in_scope
+
+    def check_expr_interpret(self, expr, gt_outputs):
+        ori_in, has_node_not_in_scope = self.get_node2value(expr.inputs)
+        if not has_node_not_in_scope:
+            expr_res = expr.interpret(*ori_in)
+            try:
+                self.check_expr_results(expr_res, gt_outputs)
+            except:
+                raise ValueError("Error occurred when checking expr: {}".format(expr))
+
+    def check_apply(self, expr, gt_outputs, opdef):
+        ori_in, has_node_not_in_scope = self.get_node2value(expr.inputs)
+        if not has_node_not_in_scope:
+            expr_res = expr.interpret(*ori_in)
+            indexs = self.check_apply_special_cases(opdef, len(gt_outputs))
+            try:
+                self.check_expr_results(expr_res, gt_outputs, indexs=indexs)
+            except:
+                raise ValueError("Error occurred when checking expr: {}".format(expr))
+
+    def check_builtin_module(self, module, expr, gt_outputs):
+        ori_in, has_node_not_in_scope = self.get_node2value(expr.inputs, start_idx=1)
+        if not has_node_not_in_scope:
+            ori_in.insert(0, module)
+            expr_res = expr.interpret(*ori_in)
+            try:
+                self.check_expr_results(expr_res, gt_outputs)
+            except:
+                raise ValueError(
+                    "{}, Error occurred when checking expr: {}".format(expr)
+                )

imperative/python/megengine/traced_module/expr.py

@@ -32,6 +32,7 @@ from .module_tracer import active_module_tracer, module_tracer
 from .node import ModuleNode, Node, NodeMixin, TensorNode
 from .pytree import ArgsIndex, TreeDef, _is_const_leaf, _is_leaf, tree_flatten
 from .serialization import _ModuleState
+from .tm_config import _exclude_from_trace, _get_expr_checker
 from .utils import _check_builtin_module_attr, _check_obj_attr, _convert_kwargs_to_args
 
 
@@ -611,6 +612,8 @@ class Apply(Expr):
             inp_nodes = [NodeMixin.get(inputs[0])]
             for i in inputs[1:]:
                 node = Constant.make(i)
+                if _get_expr_checker():
+                    active_module_tracer().checker.record_node2value(node, Tensor(i))
                 inp_nodes.append(node)
             apply_node = cls.make(opdef)
             for n in inp_nodes:
@@ -624,11 +627,17 @@ class Apply(Expr):
 
         unset_module_tracing()
         outputs = apply(opdef, *inputs)
+        outputs = list(map(Tensor, outputs))
         set_module_tracing()
 
         apply_node.add_outputs(outputs)
         for n, v in zip(apply_node.outputs, outputs):
             NodeMixin.wrap_safe(v, n)
+
+        if _get_expr_checker():
+            with _exclude_from_trace():
+                active_module_tracer().checker.check_apply(apply_node, outputs, opdef)
+
         return list(outputs)
 
 

imperative/python/megengine/traced_module/module_tracer.py

@@ -12,6 +12,7 @@ from .. import functional as F
 from ..core.tensor.array_method import ArrayMethodMixin
 from ..module import Module
 from ..module.qat import QATModule
+from .checker import TracedModuleChecker
 
 _active_module_tracer = None
 
@@ -128,6 +129,7 @@ class module_tracer:
 
     def __init__(self, wrap_fn):
         self._active_scopes = []
+        self.checker = TracedModuleChecker(self)
         self.patcher = Patcher(wrap_fn)
 
     @classmethod
@@ -142,9 +144,11 @@ class module_tracer:
 
     def push_scope(self, scope):
         self._active_scopes.append(scope)
+        self.checker.push_scope()
 
     def pop_scope(self):
         self._active_scopes.pop()
+        self.checker.pop_scope()
 
     def current_scope(self):
         if self._active_scopes:

imperative/python/megengine/traced_module/node.py

@@ -18,6 +18,8 @@ from ..core._imperative_rt.core2 import Tensor as RawTensor
 from ..module import Module
 from ..quantization.utils import QParams
 from ..tensor import Tensor
+from .module_tracer import active_module_tracer
+from .tm_config import _get_expr_checker
 from .utils import _check_obj_attr
 
 logger = get_logger(__name__)
@@ -343,6 +345,11 @@ class NodeMixin(abc.ABC):
                 if isinstance(value, NodeMixin):
                     value._record_wrapped_nodes(node)
                 setattr(value, "_NodeMixin__node", node)
+                if _get_expr_checker():
+                    if isinstance(value, RawTensor):
+                        active_module_tracer().checker.record_node2value(node, value)
+                    if isinstance(value, NodeMixin):
+                        active_module_tracer().checker.record_nodemixin(node, value)
             else:
                 assert callable(node)
                 n = node()
@@ -352,6 +359,11 @@ class NodeMixin(abc.ABC):
                 if isinstance(value, NodeMixin):
                     value._record_wrapped_nodes(n)
                 setattr(value, "_NodeMixin__node", n)
+                if _get_expr_checker():
+                    if isinstance(value, RawTensor):
+                        active_module_tracer().checker.record_node2value(n, value)
+                    if isinstance(value, NodeMixin):
+                        active_module_tracer().checker.record_nodemixin(n, value)
 
     @classmethod
     def wrap_safe(cls, value, node):
@@ -359,6 +371,11 @@ class NodeMixin(abc.ABC):
         if isinstance(value, RawTensor):
             cls._record_tensornode_property(node, value)
         setattr(value, "_NodeMixin__node", node)
+        if _get_expr_checker():
+            if isinstance(value, RawTensor):
+                active_module_tracer().checker.record_node2value(node, value)
+            if isinstance(value, NodeMixin):
+                active_module_tracer().checker.record_nodemixin(node, value)
         if isinstance(value, NodeMixin):
             value._record_wrapped_nodes(node)
 

imperative/python/megengine/traced_module/pytree.py

@@ -212,7 +212,11 @@ def tree_flatten(
     to reconstruct the pytree.
     """
     if type(values) not in SUPPORTED_TYPE:
-        assert is_leaf(values), values
+        assert is_leaf(
+            values
+        ), 'doesn\'t support {} type, MUST use "register_supported_type" method to register self-defined type'.format(
+            values
+        )
         node = LeafDef(leaf_type(values))
         if is_const_leaf(values):
             node.const_val = values

imperative/python/megengine/traced_module/tm_config.py

@@ -0,0 +1,55 @@
+import contextlib
+
+from ..core._imperative_rt.core2 import (
+    is_tracing_module,
+    set_module_tracing,
+    unset_module_tracing,
+)
+
+_enable_expr_checker = False
+_enable_default_checker = True
+
+
+def _get_expr_checker():
+    return _enable_expr_checker
+
+
+def _get_default_checker():
+    return _enable_default_checker
+
+
+def enable_expr_checker():
+    r"""Call this function to check the result of each expr during tracing."""
+    global _enable_expr_checker
+    _enable_expr_checker = True
+    _enable_default_checker = False
+
+
+def disable_default_checker():
+    r"""Call this function to disable checking the final output of the model after tracing."""
+    global _enable_default_checker
+    _enable_default_checker = False
+
+
+_enable_graph_surgery_mode = False
+
+
+def _graph_surgery_mode():
+    return _enable_graph_surgery_mode
+
+
+def _set_graph_surgery_mode(mode: bool):
+    global _enable_graph_surgery_mode
+    pre_mode = _enable_graph_surgery_mode
+    _enable_graph_surgery_mode = mode
+    return pre_mode
+
+
+@contextlib.contextmanager
+def _exclude_from_trace():
+    is_tracing = is_tracing_module()
+    if is_tracing:
+        unset_module_tracing()
+    yield
+    if is_tracing:
+        set_module_tracing()

imperative/python/megengine/traced_module/traced_module.py

@@ -36,11 +36,14 @@ from .. import get_logger
 from .. import module as M
 from ..core._imperative_rt.core2 import Tensor as RawTensor
 from ..core._imperative_rt.core2 import (
+    apply,
     is_tracing_module,
     set_module_tracing,
     unset_module_tracing,
 )
 from ..core._trace_option import set_symbolic_shape
+from ..core.ops.builtin import Copy
+from ..core.tensor.utils import isscalar, setscalar
 from ..module import Module
 from ..module import external as MExternal
 from ..module.qat import QATModule
@@ -98,6 +101,13 @@ from .serialization import (
     load_call_tensor_method_expr,
     load_functional,
 )
+from .tm_config import (
+    _exclude_from_trace,
+    _get_default_checker,
+    _get_expr_checker,
+    _graph_surgery_mode,
+    _set_graph_surgery_mode,
+)
 from .utils import (
     _check_builtin_module_attr,
     _check_obj_attr,
@@ -117,26 +127,14 @@ def _is_builtin_name(name: str) -> bool:
 
 
 def _is_leaf(node):
-    assert isinstance(node, RawTensor), "doesn't support {} in return values".format(
+    assert isinstance(
+        node, RawTensor
+    ), 'doesn\'t support {} in return values, MUST use Tensor or use "register_supported_type" method to register self-defined type'.format(
         type(node)
     )
     return isinstance(node, RawTensor)
 
 
-_enable_graph_surgery_mode = False
-
-
-def _graph_surgery_mode():
-    return _enable_graph_surgery_mode
-
-
-def _set_graph_surgery_mode(mode: bool):
-    global _enable_graph_surgery_mode
-    pre_mode = _enable_graph_surgery_mode
-    _enable_graph_surgery_mode = mode
-    return pre_mode
-
-
 def _node_to_tensor(*args, **kwargs):
     tensors = []
     nodes, tree_def = tree_flatten((args, kwargs))
@@ -1295,7 +1293,12 @@ def _wrapped_function(orig_func):
                         return orig_func(*args, **kwargs)
                     if isinstance(args[1], RawTensor):
                         node = NodeMixin.get(inputs[1])
-                        inputs[1] = copy.copy(inputs[1])
+                        is_scalar = isscalar(inputs[1])
+                        inputs[1] = apply(
+                            Copy(comp_node=inputs[1].device), Tensor(inputs[1])
+                        )[0]
+                        if is_scalar:
+                            setscalar(inputs[1])
                         # copy inputs[1] to avoid tensor and Tensor(tensor) share same m_tensor,
                         # which will cause they have same _NodeMixin__node in tracing.
                         NodeMixin.wrap_safe(inputs[1], node)
@@ -1319,6 +1322,13 @@ def _wrapped_function(orig_func):
             else:
                 outputs = None
             call_node.add_outputs(outputs)
+
+            if _get_expr_checker():
+                with _exclude_from_trace():
+                    active_module_tracer().checker.check_expr_interpret(
+                        call_node, outputs
+                    )
+
             set_module_tracing()
             return rst
         return orig_func(*args, **kwargs)
@@ -1500,6 +1510,12 @@ class TracedModuleBuilder(NodeMixin):
             unset_module_tracing()
             rst = self._mod(*args, **kwargs)
             outputs, out_def = tree_flatten(rst, is_leaf=_is_leaf)
+            if _get_expr_checker():
+                with _exclude_from_trace():
+                    tmp = self.build()
+                    active_module_tracer().checker.check_builtin_module(
+                        tmp, callnode, outputs
+                    )
             set_module_tracing()
             if self._is_builtin:
                 self._body = None
@@ -1674,7 +1690,9 @@ class TracedModuleBuilder(NodeMixin):
                     if not isinstance(mod_attr, (List, Dict, QATModule)):
                         assert mod_attr is wrapped._mod
                 else:
-                    assert mod_attr is wrapped
+                    assert (
+                        mod_attr is wrapped
+                    ), "TracedModule do not support modify attributes, please check your code."
 
                 if isinstance(wrapped, (NodeMixin, RawTensor)):
                     NodeMixin.wrap(
@@ -2469,11 +2487,23 @@ def trace_module(
                             qualname="{}.[{}]".format(net_name, "arg_{}".format(_)),
                         ),
                     )
-            builder(*args, **kwargs)
+            rst = builder(*copy.deepcopy(args), **copy.deepcopy(kwargs))
             active_module_tracer().pop_scope()
             traced_mod = builder.build()
             traced_mod.argspec = forward_argspec
             traced_mod.graph._reset_ids()
+
+            has_expr_not_check = False
+            if _get_expr_checker():
+                has_expr_not_check = (
+                    active_module_tracer().checker.check_node_not_in_scope()
+                )
+            if _get_default_checker() or has_expr_not_check:
+                with _exclude_from_trace():
+                    tm_res = traced_mod(*args, **kwargs)
+                    tm_res, _ = tree_flatten(tm_res, is_leaf=_is_leaf)
+                    rst, _ = tree_flatten(rst, is_leaf=_is_leaf)
+                    active_module_tracer().checker.check_net_outputs(tm_res, rst)
             return traced_mod
     finally:
         set_symbolic_shape(use_sym_shape)

imperative/python/megengine/traced_module/utils.py

@@ -5,16 +5,15 @@
 # Unless required by applicable law or agreed to in writing,
 # software distributed under the License is distributed on an
 # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-import collections
 import copy
 import inspect
 from collections.abc import MutableMapping, MutableSequence
 from inspect import FullArgSpec
-from typing import Callable, Dict, Iterable, List, Optional, Sequence, Type, Union
+from typing import Callable, Dict, Iterable, List, Optional, Sequence, Union
 
 from .. import get_logger
 from ..module import Module
-from ..tensor import Parameter, Tensor
+from ..tensor import Tensor
 
 logger = get_logger(__name__)
 

imperative/python/test/unit/traced_module/test_qat_module.py

@@ -109,6 +109,7 @@ def build_observered_net(net: M.Module, observer_cls):
     )
     Q.enable_observer(qat_net)
     inp = Tensor(np.random.random(size=(5, 3, 32, 32)))
+    qat_net.eval()
     qat_net(inp)
     Q.disable_observer(qat_net)
     return qat_net
@@ -116,6 +117,7 @@ def build_observered_net(net: M.Module, observer_cls):
 
 def build_fakequanted_net(net: QATModule, fakequant_cls):
     qat_net = Q.reset_qconfig(net, get_lsq_config(fakequant_cls))
+    qat_net.eval()
     return qat_net
 
 
@@ -162,6 +164,7 @@ def test_load_param():
 
     def _check_module(build_func: Callable):
         net = build_func()
+        net.eval()
         buffer = io.BytesIO()
         mge.save(net.state_dict(), buffer)
         buffer.seek(0)
@@ -185,6 +188,7 @@ def test_load_param():
 def test_qualname():
     def _check_qualname(net):
         inp = Tensor(np.random.random(size=(5, 3, 32, 32)))
+        net.eval()
         traced_net = trace_module(net, inp)
         base_qualname = traced_net.graph.qualname
         for node in traced_net.graph.nodes():