Make chunking more tolerant #24
Description
Currently, chunking requires the number of decoder layer per chunk be a multiple of the total number of decoder layers. Could lessen this restriction by simply letting the remainder be a smaller chunk.
Changes required in the converter.py module.