如何编写CUDA内核来加速python代码

几周来，我一直在学习python作为我的第一种编程语言。我决定用Numba编写一个乐透模拟。该代码在我的CPU上每秒大约250k次迭代时运行得很好。我真的很想看看它是如何在我的英伟达GPU上运行的，但我有点力不从心。如果有人能帮我一把，我将非常感激。

我的代码：

"""Importing the time module."""
import time
from numba import jit # Importing the jit function from the numba module.
import numpy as np # Importing the numpy library and giving it the alias np.

start = time.time()


@jit(nopython=True)
def lotto_numbers():
    """
    The lotto_numbers function generates a list of six random numbers between 1 and 59.
    The function will not return any number that is zero, but it will return any other number.


    :return: A sorted numpy array of 6 unique random numbers between 1 and 59
    """
    generated = False
    while not generated:
        balls = np.random.choice(59, 6, replace=False)
        balls.sort()
        temp = np.where(balls == 0)
        (check_for_zero,) = temp[0].shape
        if check_for_zero == 0:
            generated = True
    return balls


your_numbers = lotto_numbers()  # optionally choose 6 numbers (not yet implemented).
print(f"Your numbers are: {your_numbers}")

MATCH = False
GAME_COUNT = 0
while not MATCH:
    tmp = lotto_numbers()
    cmp = tmp == your_numbers
    GAME_COUNT += 1
    if cmp.all():
        MATCH = True
        TOTAL_COST = GAME_COUNT * 2
        TOTAL_WEEKS = GAME_COUNT / 2
        TOTAL_YEARS = round(TOTAL_WEEKS / 52, 1)
        end = time.time()
        TOTAL_TIME = end - start
        GPS = GAME_COUNT / TOTAL_TIME

        print(f"\n\nWINNER!!! after {GAME_COUNT:,} games")
        print(f"\ttotal cost: £{TOTAL_COST:,}")
        print(f"\tweeks until win: {TOTAL_WEEKS:,} weeks")
        print(f"\tyears until win: {TOTAL_YEARS:,} years")
        print(
            f"\texecution time: {TOTAL_TIME:,.2f} seconds at {GPS:,.2f} games per second")

    if GAME_COUNT % 10000 == 0:
        print(f"Game : {GAME_COUNT:,}", end='\r')

我知道我需要编写一个CUDA内核来针对GPU。我想我应该能够运行float16，因为数字并不复杂。此外，@vectorize似乎很重要。但是，老实说，我在踩水。