fix(mge/quantization): use list rather than generator to modify submodules

GitOrigin-RevId: b5a3851330e7ffeb2aa9ad1a42e8ade8a0817506

python_module/megengine/quantization/quantize.py

@@ -65,9 +65,9 @@ def quantize(module: Module, inplace=True):
     def is_qat(mod: Module):
         return isinstance(mod, qat_modules)
 
-    # no need to pass prefix and get pure key of parent Module.
-    for key, submodule, parent in module._flatten(
-        with_key=True, with_parent=True, predicate=is_qat
+    # must use list to avoid replacement influencing successor modules
+    for key, submodule, parent in list(
+        module._flatten(with_key=True, with_parent=True, predicate=is_qat)
     ):
         new_mod = _qat2quantized_dict[type(submodule)].from_qat_module(submodule)
         if isinstance(parent, Float.Sequential):
@@ -100,9 +100,9 @@ def quantize_qat(
     def is_quantable(mod: Module):
         return isinstance(mod, quantable_modules)
 
-    # no need to pass prefix and get pure key of parent Module.
-    for key, submodule, parent in module._flatten(
-        with_key=True, with_parent=True, predicate=is_quantable
+    # must use list to avoid replacement influencing successor modules
+    for key, submodule, parent in list(
+        module._flatten(with_key=True, with_parent=True, predicate=is_quantable)
     ):
         # only convert top quantable module.
         if is_quantable(parent):