add changelog and fix test

CHANGELOG.md

@@ -735,6 +735,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed to avoid common hook warning if no hook is overridden ([#12131](https://github.com/PyTorchLightning/pytorch-lightning/pull/12131))
 
 
+- Fixed an issue that metric states are incorretly saved and restored on checkpointing when `_fault_tolerant_training() = False` ([#12236](https://github.com/PyTorchLightning/pytorch-lightning/pull/12236))
+
+
 ## [1.5.10] - 2022-02-08
 
 ### Fixed

tests/callbacks/test_quantization.py

@@ -16,6 +16,7 @@
 
 import pytest
 import torch
+from torchmetrics import Metric
 from torchmetrics.functional import mean_absolute_percentage_error as mape
 
 from pytorch_lightning import seed_everything, Trainer
@@ -243,5 +244,7 @@ def test_quantization_val_test_predict(tmpdir):
 
     expected_state_dict = expected_qmodel.state_dict()
     for key, value in val_test_predict_qmodel.state_dict().items():
-        expected_value = expected_state_dict[key]
-        assert torch.allclose(value, expected_value)
+        # validate and test will affect metric module state, other modules should be comparable.
+        if not isinstance(val_test_predict_qmodel.get_submodule(key.split(".")[0]), Metric):
+            expected_value = expected_state_dict[key]
+            assert torch.allclose(value, expected_value)