Odd result using multinomial num_samples 1 vs larger sizes

[NOTE: I started on punch cards in 1981 and have written well over a million lines of application code in my career.)

I’m seeing the exact same seed number producing the exact same output from the generator for any values I tested (I actually used 2,3,4,5,10,20 - all produced the correct [according to the video] output); 13.

However, when I use num_samples = 1, the generator produces the exact same output (0.7081), but that results in a result of index 10 instead of 13 which, as can be seen in the cumulative tensor, should be a value between 0.4824 and 0.5749.

I just can’t determine what could be going on.