Hm, I'd separate these into "chunks" that are each a file, and worry those apart later. Something like this:

open INFILE, '<', 'infile.txt' or die("Can't read: $!");

# each 'line' is a chunk with 'filename =' delimiting.
local $/ = 'filename =';

while (<INFILE>) {
   s/^s\+//s; # remove starting whitespace.
   
   # use a regex capture -- slow but simple.
   my ($filename, $data) = ($1, $2) if m/^(\w+)\s+pred=\s+(.*?)\s+$/s;

   open my $OUT, '>', $filname or die("Can't write $filename: $!");
   foreach ( split(/\s+|\n/, $data) ) {
      next unless m/^[-.0-9][0-9.]+/; #skip any empty or malformed lin
+es
      print $OUT $_;
   }
}
[download]

Untested, but the principle is sound and somewhat reusable.