FIx #7

Fix #7

This was a fun one, had a missing len() which caused relation candidates to be 0 when candidates are trimmed due to max_length. Due to an older version of pytorch having issues with 0 dimensional tensors with einsum, it cause issues to some people on Colab.

relik/reader/data/relik_reader_re_data.py

@@ -730,7 +730,7 @@ def flip_cands(flip_candidates, candidates):
 
                     candidates_encoding_result = (
                         candidates_encoding_result[:i]
-                        + candidates_encoding_result[len(sample.span_candidates) : i]
+                        + candidates_encoding_result[len(sample.span_candidates) : len(sample.span_candidates) + i]
                     )
                     if len(self.special_symbols_types) > 0:
                         candidates_entities_symbols = candidates_entities_symbols[:i]