fairseq/tests/test_ema.py

# Copyright (c) Facebook, Inc. and its affiliates.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.

import unittest
from copy import deepcopy
from dataclasses import dataclass
import pytest
from typing import Optional
from unittest.mock import patch

import torch

from fairseq.models.ema import EMA


class DummyModule(torch.nn.Module):
    def __init__(self) -> None:
        """LightningModule for testing purposes

        Args:
            epoch_min_loss_override (int, optional): Pass in an epoch that will be set to the minimum
                validation loss for testing purposes (zero based). If None this is ignored. Defaults to None.
        """
        super().__init__()
        self.layer = torch.nn.Linear(in_features=32, out_features=2)
        self.another_layer = torch.nn.Linear(in_features=2, out_features=2)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.layer(x)
        return self.another_layer(x)


@dataclass
class EMAConfig(object):
    ema_decay: float = 0.99
    ema_start_update: int = 0
    ema_fp32: bool = False
    ema_seed_model: Optional[str] = None
    ema_update_freq: int = 1


class TestEMA(unittest.TestCase):
    def assertTorchAllClose(self, x, y, atol=1e-8, rtol=1e-5, msg=None):
        diff = x.float() - y.float()
        diff_norm = torch.norm(diff)
        other_norm = torch.norm(y.float())

        if msg is None:
            msg = "|input - other| > {} + {} * |other|".format(atol, rtol)

        self.assertLessEqual(
            diff_norm,
            atol + rtol * other_norm,
            msg=msg,
        )

    def test_ema(self):
        model = DummyModule()
        optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
        state = deepcopy(model.state_dict())
        config = EMAConfig()
        ema = EMA(model, config)

        # set decay
        ema._set_decay(config.ema_decay)
        self.assertEqual(ema.get_decay(), config.ema_decay)

        # get model
        self.assertEqual(ema.get_model(), ema.model)

        # Since fp32 params is not used, it should be of size 0
        self.assertEqual(len(ema.fp32_params), 0)

        # EMA step
        x = torch.randn(32)
        y = model(x)
        loss = y.sum()
        loss.backward()
        optimizer.step()

        ema.step(model)

        ema_state_dict = ema.get_model().state_dict()

        for key, param in model.state_dict().items():
            prev_param = state[key]
            ema_param = ema_state_dict[key]

            if "version" in key:
                # Do not decay a model.version pytorch param
                continue
            self.assertTorchAllClose(
                ema_param,
                config.ema_decay * prev_param + (1 - config.ema_decay) * param,
            )

        # Since fp32 params is not used, it should be of size 0
        self.assertEqual(len(ema.fp32_params), 0)

        # Load EMA into model
        model2 = DummyModule()
        ema.reverse(model2)

        for key, param in model2.state_dict().items():
            ema_param = ema_state_dict[key]
            self.assertTrue(torch.allclose(ema_param, param))

        # Check that step_internal is called once
        with patch.object(ema, "_step_internal", return_value=None) as mock_method:
            ema.step(model)
            mock_method.assert_called_once_with(model, None)

    def _test_ema_start_update(self, updates):
        model = DummyModule()
        optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
        state = deepcopy(model.state_dict())
        config = EMAConfig(ema_start_update=1)
        ema = EMA(model, config)

        # EMA step
        x = torch.randn(32)
        y = model(x)
        loss = y.sum()
        loss.backward()
        optimizer.step()

        ema.step(model, updates=updates)
        ema_state_dict = ema.get_model().state_dict()

        self.assertEqual(ema.get_decay(), 0 if updates == 0 else config.ema_decay)

        for key, param in model.state_dict().items():
            ema_param = ema_state_dict[key]
            prev_param = state[key]

            if "version" in key:
                # Do not decay a model.version pytorch param
                continue
            if updates == 0:
                self.assertTorchAllClose(
                    ema_param,
                    param,
                )
            else:
                self.assertTorchAllClose(
                    ema_param,
                    config.ema_decay * prev_param + (1 - config.ema_decay) * param,
                )

        # Check that step_internal is called once
        with patch.object(ema, "_step_internal", return_value=None) as mock_method:
            ema.step(model, updates=updates)
            mock_method.assert_called_once_with(model, updates)

    def test_ema_before_start_update(self):
        self._test_ema_start_update(updates=0)

    def test_ema_after_start_update(self):
        self._test_ema_start_update(updates=1)

    def test_ema_fp32(self):
        dtype = torch.float

        model = DummyModule().to(dtype)
        optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
        state = deepcopy(model.state_dict())
        config = EMAConfig(ema_fp32=True)
        ema = EMA(model, config)

        x = torch.randn(32)
        y = model(x.to(dtype))
        loss = y.sum()
        loss.backward()
        optimizer.step()

        ema.step(model)

        for key, param in model.state_dict().items():
            prev_param = state[key]
            ema_param = ema.get_model().state_dict()[key]

            if "version" in key:
                # Do not decay a model.version pytorch param
                continue
            self.assertIn(key, ema.fp32_params)

            # EMA update is done in fp32, and hence the EMA param must be
            # closer to the EMA update done in fp32 than in fp16.
            self.assertLessEqual(
                torch.norm(
                    ema_param.float()
                    - (
                        config.ema_decay * prev_param.float()
                        + (1 - config.ema_decay) * param.float()
                    )
                    .to(dtype)
                    .float()
                ),
                torch.norm(
                    ema_param.float()
                    - (
                        config.ema_decay * prev_param + (1 - config.ema_decay) * param
                    ).float()
                ),
            )
            self.assertTorchAllClose(
                ema_param,
                (
                    config.ema_decay * prev_param.float()
                    + (1 - config.ema_decay) * param.float()
                ).to(dtype),
            )

    @pytest.mark.skipif(
        not torch.cuda.is_available(),
        reason="CPU no longer supports Linear in half precision",
    )
    def test_ema_fp16(self):
        model = DummyModule().cuda().half()
        optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
        state = deepcopy(model.state_dict())
        config = EMAConfig(ema_fp32=False)
        ema = EMA(model, config)

        # Since fp32 params is not used, it should be of size 0
        self.assertEqual(len(ema.fp32_params), 0)

        x = torch.randn(32).cuda()
        y = model(x.half())
        loss = y.sum()
        loss.backward()
        optimizer.step()

        ema.step(model)

        for key, param in model.state_dict().items():
            prev_param = state[key]
            ema_param = ema.get_model().state_dict()[key]

            if "version" in key:
                # Do not decay a model.version pytorch param
                continue

            # EMA update is done in fp16, and hence the EMA param must be
            # closer to the EMA update done in fp16 than in fp32.
            self.assertLessEqual(
                torch.norm(
                    ema_param.float()
                    - (
                        config.ema_decay * prev_param + (1 - config.ema_decay) * param
                    ).float()
                ),
                torch.norm(
                    ema_param.float()
                    - (
                        config.ema_decay * prev_param.float()
                        + (1 - config.ema_decay) * param.float()
                    )
                    .half()
                    .float()
                ),
            )
            self.assertTorchAllClose(
                ema_param,
                config.ema_decay * prev_param + (1 - config.ema_decay) * param,
            )

        # Since fp32 params is not used, it should be of size 0
        self.assertEqual(len(ema.fp32_params), 0)


if __name__ == "__main__":
    unittest.main()