Check this

In brief: Unordered set and unordered map use hash table (imagine it as a large array) and for each item it uses some algorithm to find the index it should be stored in (which usually works in $$$O(1)$$$), if something is already stored at that index, for simplicity let's say it puts it next to it, if something is next to it, it finds the next empty index and inserts it there, this is known as collision, now if the algorithm determines that most items should be stored in the same index what will happen? it will keep searching for the next empty one which will lead the complexity to increase from $$$O(1)$$$ to $$$O(n)$$$, this rarely happen unless a test case is specifically written to cause this (which someone will mostly do to hack your solution if the author didn't), so for codeforces contests just avoid using them unless you are sure no collisions can happen.

The reason you get TLE when using an unordered_map is that, although the average complexity of insert() and find() is O(1), their worse case complexity is O(n). This makes it possible to hack a solution that uses unordered_map in some cases.

The complexity of insert() and find() on a map is O(logn), which is efficient enough most of the time. I'm pretty sure I never got TLE because of using a map instead of an unordered_map (Maybe I was just lucky, though. Did that ever happpen to you?), so I don't think there's a reason not to prefer it.