test/jit/test_module_apis.py

import torch
import os
import sys
from torch.testing._internal.jit_utils import JitTestCase
from typing import Dict, Any, List

# Make the helper files in test/ importable
pytorch_test_dir = os.path.dirname(os.path.dirname(os.path.realpath(__file__)))
sys.path.append(pytorch_test_dir)

if __name__ == '__main__':
    raise RuntimeError("This test file is not meant to be run directly, use:\n\n"
                       "\tpython test/test_jit.py TESTNAME\n\n"
                       "instead.")

class TestModuleAPIs(JitTestCase):
    def test_default_state_dict_methods(self):
        """Tests that default state dict methods are automatically available"""

        class DefaultStateDictModule(torch.nn.Module):
            def __init__(self):
                super().__init__()
                self.conv = torch.nn.Conv2d(6, 16, 5)
                self.fc = torch.nn.Linear(16 * 5 * 5, 120)

            def forward(self, x):
                x = self.conv(x)
                x = self.fc(x)
                return x

        m1 = torch.jit.script(DefaultStateDictModule())
        m2 = torch.jit.script(DefaultStateDictModule())
        state_dict = m1.state_dict()
        m2.load_state_dict(state_dict)

    def test_customized_state_dict_methods(self):
        """Tests that customized state dict methods are in effect"""

        class CustomStateDictModule(torch.nn.Module):
            def __init__(self):
                super().__init__()
                self.conv = torch.nn.Conv2d(6, 16, 5)
                self.fc = torch.nn.Linear(16 * 5 * 5, 120)
                self.customized_save_state_dict_called: bool = False
                self.customized_load_state_dict_called: bool = False

            def forward(self, x):
                x = self.conv(x)
                x = self.fc(x)
                return x

            @torch.jit.export
            def _save_to_state_dict(self, destination: Dict[str, torch.Tensor],
                                    prefix: str, keep_vars: bool):
                self.customized_save_state_dict_called = True
                return {"dummy": torch.ones(1)}

            @torch.jit.export
            def _load_from_state_dict(self,
                                      state_dict: Dict[str, torch.Tensor],
                                      prefix: str, local_metadata: Any,
                                      strict: bool, missing_keys: List[str],
                                      unexpected_keys: List[str],
                                      error_msgs: List[str]):
                self.customized_load_state_dict_called = True
                return

        m1 = torch.jit.script(CustomStateDictModule())
        self.assertFalse(m1.customized_save_state_dict_called)
        state_dict = m1.state_dict()
        self.assertTrue(m1.customized_save_state_dict_called)

        m2 = torch.jit.script(CustomStateDictModule())
        self.assertFalse(m2.customized_load_state_dict_called)
        m2.load_state_dict(state_dict)
        self.assertTrue(m2.customized_load_state_dict_called)

    def test_submodule_customized_state_dict_methods(self):
        """Tests that customized state dict methods on submodules are in effect"""

        class CustomStateDictModule(torch.nn.Module):
            def __init__(self):
                super().__init__()
                self.conv = torch.nn.Conv2d(6, 16, 5)
                self.fc = torch.nn.Linear(16 * 5 * 5, 120)
                self.customized_save_state_dict_called: bool = False
                self.customized_load_state_dict_called: bool = False

            def forward(self, x):
                x = self.conv(x)
                x = self.fc(x)
                return x

            @torch.jit.export
            def _save_to_state_dict(self, destination: Dict[str, torch.Tensor],
                                    prefix: str, keep_vars: bool):
                self.customized_save_state_dict_called = True
                return {"dummy": torch.ones(1)}

            @torch.jit.export
            def _load_from_state_dict(self,
                                      state_dict: Dict[str, torch.Tensor],
                                      prefix: str, local_metadata: Any,
                                      strict: bool, missing_keys: List[str],
                                      unexpected_keys: List[str],
                                      error_msgs: List[str]):
                self.customized_load_state_dict_called = True
                return

        class ParentModule(torch.nn.Module):
            def __init__(self):
                super().__init__()
                self.sub = CustomStateDictModule()

            def forward(self, x):
                return self.sub(x)

        m1 = torch.jit.script(ParentModule())
        self.assertFalse(m1.sub.customized_save_state_dict_called)
        state_dict = m1.state_dict()
        self.assertTrue(m1.sub.customized_save_state_dict_called)

        m2 = torch.jit.script(ParentModule())
        self.assertFalse(m2.sub.customized_load_state_dict_called)
        m2.load_state_dict(state_dict)
        self.assertTrue(m2.sub.customized_load_state_dict_called)