Is torch.random with manual_seed less random within a jupyter code block

[innovinitylabs]

in question 4 of the 00_pytorch_fundamentals_exercises

I set manual_seed each time before generating a random tensor but didn't got the answer as required.
so i didn't set manual_seed when creating (1,7) tensor. but surprisingly i got the required answer.
but not setting manual_seed before creating a random tensor should give unpredictable resources right?

so i visualise, visualise, visualised like in the attached image and found that setting a manual_seed makes less random within same code cell.

heres my code below

edit:

after creating this i figured that the pattern of values remains same in the code block. like the third random tensor always starts with 0.2666

[innovinitylabs]

it's not a problem with notebooks, I just overlooked the fact that

• for each call, the random value changes even if the seed is set its just that its predictable standard ones for given seed

• when i rerun the cell where manual_seed is set it's always n & n+1 call of the random function, so no matter how many times i call it remains the same, but when i call the random function in the next cell, every call increases the number of times it was called and so it kept changing (which i perceived as not stable).