chore: restore cifar_utils.py

use_case_examples/cifar/cifar_brevitas_finetuning/cifar_utils.py

@@ -91,30 +91,6 @@
             ]
         ),
     },
-    "MNIST": {
-        "dataset": datasets.MNIST,
-        "mean": (0.5),
-        "std": (0.5),
-        "train_transform": transforms.Compose(
-            [
-                transforms.Pad(1, padding_mode="edge"),
-                transforms.ToTensor(),
-                transforms.Normalize((0.5,), (0.5,)),
-                # transforms.RandomRotation(5, fill=(1,)),
-                transforms.GaussianBlur(kernel_size=(3, 3)),
-                # transforms.RandomHorizontalFlip(0.5),
-                # transforms.Resize(20),
-            ]
-        ),
-        "test_transform": transforms.Compose(
-            [
-                transforms.Pad(1, padding_mode="edge"),
-                transforms.ToTensor(),
-                transforms.Normalize((0.5,), (0.5,)),
-                # transforms.Resize(20),
-            ]
-        ),
-    },
 }
 
 
@@ -164,14 +140,11 @@ def get_dataloader(
         Tuple[DataLoader, DataLoader]: Training and test data loaders.
     """
 
-    g = None
-
-    if param["seed"]:
-        g = torch.Generator()
-        g.manual_seed(param["seed"])
-        np.random.seed(param["seed"])
-        torch.manual_seed(param["seed"])
-        random.seed(param["seed"])
+    g = torch.Generator()
+    g.manual_seed(param["seed"])
+    np.random.seed(param["seed"])
+    torch.manual_seed(param["seed"])
+    random.seed(param["seed"])
 
     max_examples = param.get("dataset_size", None)
     train_dataset = get_torchvision_dataset(
@@ -360,10 +333,8 @@ def train(
         nn.Module: the trained model.
     """
 
-    if param["seed"]:
-
-        torch.manual_seed(param["seed"])
-        random.seed(param["seed"])
+    torch.manual_seed(param["seed"])
+    random.seed(param["seed"])
 
     model = model.to(device)
 
@@ -372,10 +343,6 @@ def train(
         optimizer, milestones=param["milestones"], gamma=param["gamma"]
     )
 
-    # Save the state_dict
-    dir = Path(".") / param["dir"] / param["training"]
-    dir.mkdir(parents=True, exist_ok=True)
-
     # To avoid breaking up the tqdm bar
     with tqdm(total=param["epochs"], file=sys.stdout) as pbar:
 
@@ -428,17 +395,12 @@ def train(
                 )
                 pbar.update(step)
 
-            torch.save(
-                model.state_dict(),
-                f"{dir}/{param['dataset_name']}_{param['training']}_state_dict.pt",
-            )
-
-    print("Save in:", f"{dir}/{param['dataset_name']}_{param['training']}_state_dict.pt")
-
+    # Save the state_dict
+    dir = Path(".") / param["dir"] / param["training"]
+    dir.mkdir(parents=True, exist_ok=True)
     torch.save(
         model.state_dict(), f"{dir}/{param['dataset_name']}_{param['training']}_state_dict.pt"
     )
-    import pickle as pkl
 
     with open(f"{dir}/{param['dataset_name']}_history.pkl", "wb") as f:
         pkl.dump(param, f)
@@ -479,24 +441,12 @@ def torch_inference(
     return np.mean(np.vstack(correct), dtype="float64")
 
 
-def fhe_compatibility(
-    model: Callable,
-    data: DataLoader,
-    rounding_threshold_bits: Optional[int] = None,
-    show_mlir: bool = False,
-    output_onnx_file: str = "test.onnx",
-) -> Callable:
+def fhe_compatibility(model: Callable, data: DataLoader) -> Callable:
     """Test if the model is FHE-compatible.
 
     Args:
         model (Callable): The Brevitas model.
         data (DataLoader): The data loader.
-        rounding_threshold_bits (Optiona[int]): if not None, every accumulators in the model are
-            rounded down to the given bits of precision.
-        show_mlir (bool): if set, the MLIR produced by the converter and which is going
-            to be sent to the compiler backend is shown on the screen, e.g., for debugging or demo.
-        output_onnx_file (str): temporary file to store ONNX model. If None a temporary file
-            is generated.
 
     Returns:
         Callable: Quantized model.
@@ -506,9 +456,8 @@ def fhe_compatibility(
         model.to("cpu"),
         # Training
         torch_inputset=data,
-        show_mlir=show_mlir,
-        output_onnx_file=output_onnx_file,
-        rounding_threshold_bits=rounding_threshold_bits,
+        show_mlir=False,
+        output_onnx_file="test.onnx",
     )
 
     return qmodel