Upload new GPTQs with varied parameters

Update tokenizer_config.json to match other models, like Vicuna - the Koala conversion process does not fill out this file, but based on the code I think these special tokens should be defined.

Push model