[Feature] Custom stopping word list
ElvishScout opened this issue · comments
Feature Request
On some llama models text generation won't stop after hitting the <|im_end|> tag. There is a workaround by adding <|im_end|> to the stopping word list on other platforms such as llama-cpp-python.
I wonder how this problem can be fixed on GPT4All.
Related stop token issue, which stop word list will be a workaround for: #2167 (comment)
Competes with #2439
Closing in favor of #2439