avoiding duplicate instances of an object with random variables

First, adjust your class to take the x and y coordinates as initialization parameters.

class Target:
    def __init__(self, x, y):  
        self.x = x
        self.y = y 

Then you could generate a list of all valid coordinate pairs.

xy = [(x, y) for x in range(1, 10) for y in range(1, 10)]

Then take a random sample of size 10 without replacement.

xy_sample = random.sample(xy, 10)

Then initialize 10 instances of your class.

targets = [Target(x, y) for x, y in xy_sample]

Caveat: if the xy-space is huge then just keeping track of the generated x, y pairs in a set and rerolling the randomly generated numbers in the unlikely case that you hit a duplicate may be more feasible.

Assuming your class constructor's interface is set and you can't change it (or don't want to), so the randomization part needs to be inside __init__. And also assuming that just "assuring" there are no dupes in a list of Targets is primarily what you want, without any further assumptions made on what you're exactly trying to achieve.

Then, as a first step, you would require for the Targets to be comparable to each other, based on their x and y values.This can be implemented in the __eq__ method:

def __eq__(self, other):
     return self.x == other.x and self.y == other.y

Alongside, you need to create __hash__, which is required to be equal for objects that are equal based on __eq__, and also because reimplementing __eq__ resets the default __hash__ so Target wouldn't be hashable without:

def __hash__(self):
     return hash((self.x, self.y))  # just use the tuple's hash

Now you can compare Targets with one another based on their (x,y) data:

Target() == Target()

This worked before, too, but would return if the two targets are the same object, which may be False even with the same x and y.

With this as the basis, you can now assert there are no dupes in your list of targets

targets = [Target() for in range(10)]

by turning it into a set, which removes dupes based on equality based on __eq__, and then comparing the length of the result:

assert len(targets) == len(set(targets))

This is may not be what makes the most sense to do, but a direct answer to your question nonetheless.

This uses a generator that remebers the old choices.
Carefull: The function can turn into an infinity loop if you create more classes of Target than there are available combinations...

import numpy as np
class CustomGenerator(object):
    def __init__(self, lower_bound: int, upper_bound: int):
        self.old_values = []
        self.lower_bound = lower_bound
        self.upper_bound = upper_bound
    def __iter__(self):
        return self
    def __next__(self):
        return self.next()
    def next(self):
        while True:
            new_vals = tuple(np.random.randint(self.lower_bound, self.upper_bound, 2))
            if new_vals not in self.old_values:
                break
        return new_vals
gen = CustomGenerator(0, 100)
class Target:
    def __init__(self, gen):  
        self.x, self.y = next(gen)
    def __repr__(self) -> str:
        return f"{self.x}, {self.y}"
t = Target(gen)
print(t)