improve the implicit conversion rule when there are float Number and other type Number

mindspore/ccsrc/frontend/operator/composite/do_signature.cc

@@ -143,7 +143,7 @@ TypeId GetMaxTypeId(const std::vector<TypePtr> &input_types, std::vector<size_t>
     }
   }
   if (max_type_id != kNumberTypeFloat16 && max_type_id != kNumberTypeFloat32 && max_type_id != kNumberTypeFloat64 &&
-      has_scalar_float32) {
+      max_type_id != kTypeUnknown && has_scalar_float32) {
     max_type_id = kNumberTypeFloat32;
   }
   return max_type_id;

mindspore/ccsrc/pipeline/pynative/pynative_execute.cc

@@ -224,7 +224,7 @@ std::map<SignatureEnumDType, TypeId> GetDstType(const py::tuple &py_args,
       }
     }
     if (max_type != TypeId::kNumberTypeFloat16 && max_type != TypeId::kNumberTypeFloat32 &&
-        max_type != TypeId::kNumberTypeFloat64 && has_float) {
+        max_type != TypeId::kNumberTypeFloat64 && max_type != TypeId::kTypeUnknown && has_float) {
       max_type = TypeId::kNumberTypeFloat32;
     }
     if (max_type == TypeId::kNumberTypeUInt8 && has_int8) {