Technically, any odd number of backslashes, as described in the documentation.
>>> r'\'
File "<stdin>", line 1
r'\'
^
SyntaxError: EOL while scanning string literal
>>> r'\\'
'\\\\'
>>> r'\\\'
File "<stdin>", line 1
r'\\\'
^
SyntaxError: EOL while scanning string literal
It seems like the parser could just treat backslashes in raw strings as regular characters (isn't that what raw strings are all about?), but I'm probably missing something obvious.
The reason is explained in the part of that section which I highlighted in bold:
String quotes can be escaped with a backslash, but the backslash remains in the string; for example,
r"\""
is a valid string literal consisting of two characters: a backslash and a double quote;r"\"
is not a valid string literal (even a raw string cannot end in an odd number of backslashes). Specifically, a raw string cannot end in a single backslash (since the backslash would escape the following quote character). Note also that a single backslash followed by a newline is interpreted as those two characters as part of the string, not as a line continuation.
So raw strings are not 100% raw, there is still some rudimentary backslash-processing.