[1/N] Refactored AutoQuantizeSearcher to _AutoQuantizeBaseSearcher & AutoQuantizeGradientSearcher; seperated quant modules and score modules

Signed-off-by: realAsma <[email protected]>

minor

Signed-off-by: realAsma <[email protected]>

minor

Signed-off-by: realAsma <[email protected]>

chery-picked some relevant changes

Signed-off-by: realAsma <[email protected]>

updated docs; code clean up

Signed-off-by: realAsma <[email protected]>

clean ups

Signed-off-by: realAsma <[email protected]>

clean ups

Signed-off-by: realAsma <[email protected]>

Remove torch.compile decorator to fix ONNX unittests

Signed-off-by: realAsma <[email protected]>

Author: Asma Kuriparambil Thekkumpate

modelopt/torch/opt/hparam.py

@@ -48,7 +48,7 @@ def __eq__(self, other) -> bool:
 class Hparam:
     """A base hyperparameter of a DynamicModule.
 
-    An example of such a Hparam could be an hparam with identity dependencies.
+    Keeps track of hyperparameter values and their importance, which can be used for search algorithms.
     """
 
     Importance = Union[torch.Tensor, None]  # noqa: UP007
@@ -249,10 +249,14 @@ def _enforce_order(self, order: torch.Tensor | None = None) -> None:
             order = order.cpu()
         self._slice_order = order
 
+    @property
+    def attrs(self) -> list[str]:
+        """Return the attributes of the hparam for repr."""
+        return ["choices", "active", "original"]
+
     def __repr__(self) -> str:
         """Return string representation with relevant properties of the class."""
-        attrs = ["choices", "active", "original"]
-        return f"{type(self).__name__}({', '.join(f'{x}={getattr(self, x)}' for x in attrs)})"
+        return f"{type(self).__name__}({', '.join(f'{x}={getattr(self, x)}' for x in self.attrs)})"
 
     def __iand__(self, hp: "Hparam"):
         """Merge another hparam into self."""