Fix broken EMA in fairseq

Summary: EMA broken since D33649708 (995c204) due to indentation error.

Reviewed By: cruvadom

Differential Revision: D33809223

fbshipit-source-id: c6c4d0d327443bfea787817040e1832eef0f50e4

fairseq/models/ema/ema.py

@@ -185,11 +185,11 @@ def step(self, new_model, updates=None):
             self._set_decay(
                 0 if updates < self.config.ema_start_update else self.config.ema_decay
             )
-            if updates is not None and self.config.ema_update_freq > 1:
-                self.update_freq_counter += 1
-                if self.update_freq_counter >= self.config.ema_update_freq:
-                    self._step_internal(new_model, updates)
-                    self.update_freq_counter = 0
+        if self.config.ema_update_freq > 1:
+            self.update_freq_counter += 1
+            if self.update_freq_counter >= self.config.ema_update_freq:
+                self._step_internal(new_model, updates)
+                self.update_freq_counter = 0
         else:
             self._step_internal(new_model, updates)
 

tests/test_ema.py

@@ -4,6 +4,7 @@
 # LICENSE file in the root directory of this source tree.
 
 import unittest
+from unittest.mock import patch
 from copy import deepcopy
 from dataclasses import dataclass
 from typing import Optional
@@ -36,9 +37,10 @@ class EMAConfig(object):
     ema_start_update: int = 0
     ema_fp32: bool = False
     ema_seed_model: Optional[str] = None
+    ema_update_freq: int = 1
 
 
-class TestEMAGPU(unittest.TestCase):
+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)
@@ -104,6 +106,63 @@ def test_ema(self):
             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):
         model = DummyModule().half()
         optimizer = torch.optim.SGD(model.parameters(), lr=0.01)