[fbsync] Add a test that compares the output of our quantized models against expected cached values (#4597)

Summary:
* adding tests to check output of quantized models

* adding test quantized model weights

* merge test_new_quantized_classification_model with test_quantized_classification_model

* adding skipif removed by mistake

* addressing comments from PR

* removing unused argument

* fixing lint errors

* changing model to eval model and updating weights

* Update test/test_models.py

* enforce single test in circleci

* changing random seed

* updating weights for new seed

* adding missing empty line

* try 128 random seed

* try 256 random seed

* try 16 random seed

* disable inception_v3 input/output quantization tests

* removing ModelTester.test_inception_v3_quantized_expect.pkl

* reverting temporary ci run_test.sh changes

Reviewed By: fmassa

Differential Revision: D31649962

fbshipit-source-id: 35a0cb4d8d3564c88dabc09e750d5ad0a281431a

Co-authored-by: Vasilis Vryniotis <[email protected]>
Co-authored-by: Vasilis Vryniotis <[email protected]>

Author: 2 people

test/expect/ModelTester.test_googlenet_quantized_expect.pkl

@@ -220,6 +220,11 @@ def _check_input_backprop(model, inputs):
     "maskrcnn_resnet50_fpn",
 )
 
+# The tests for the following quantized models are flaky possibly due to inconsistent
+# rounding errors in different platforms. For this reason the input/output consistency
+# tests under test_quantized_classification_model will be skipped for the following models.
+quantized_flaky_models = ("inception_v3",)
+
 
 # The following contains configuration parameters for all models which are used by
 # the _test_*_model methods.
@@ -687,7 +692,9 @@ def test_video_model(model_name, dev):
 )
 @pytest.mark.parametrize("model_name", get_available_quantizable_models())
 def test_quantized_classification_model(model_name):
+    set_rng_seed(0)
     defaults = {
+        "num_classes": 5,
         "input_shape": (1, 3, 224, 224),
         "pretrained": False,
         "quantize": True,
@@ -697,8 +704,15 @@ def test_quantized_classification_model(model_name):
 
     # First check if quantize=True provides models that can run with input data
     model = torchvision.models.quantization.__dict__[model_name](**kwargs)
+    model.eval()
     x = torch.rand(input_shape)
-    model(x)
+    out = model(x)
+
+    if model_name not in quantized_flaky_models:
+        _assert_expected(out, model_name + "_quantized", prec=0.1)
+        assert out.shape[-1] == 5
+        _check_jit_scriptable(model, (x,), unwrapper=script_model_unwrapper.get(model_name, None))
+        _check_fx_compatible(model, x)
 
     kwargs["quantize"] = False
     for eval_mode in [True, False]: