The results of topic modeling on colab are unstable.

[ZstuPsyGzy]

After using GPU for topic modeling on colab, it was found that even if a fixed random seed was used, when the same data was used for topic modeling again the next day, different topics would be obtained (for example: the quantity decreased, from 20 to 2, or 20 to 15).
Will the results of bertopic modeling vary with different Gpus (could it be a problem of random seed generation at the bottom layer?) ? This issue was not detected when using the CPU before. It would be highly appreciated if anyone knew the reason or the solution.