[Chain] Summarize returns empty result when intermediate documents contain hashtag #

[vertinski]

Summarize returns empty result when intermediate documents contain hashtag #. Otherwise using different prompt which doesn't generate hashtags all is cool.

prompt_template = """Write a tweet from the following text:
{text}
"""

PROMPT = PromptTemplate (template=prompt_template, input_variables=["text"])
chain = load_summarize_chain (llm, chain_type="map_reduce", map_prompt=PROMPT, combine_prompt=PROMPT)
chain.run (docs)
''

[hwchase17]

hmm its not giving me any issues (see above)

do you want to create a google collab notebook with an example so i can replicate there?

[vertinski]

here's the notebook:

https://github.com/vertinski/langchain_bugtest/blob/main/langchain_test_001.ipynb

I think it might be the davinci-003 issue, because often it returns inconsistent results....
(also sorry for posting notebook on github!)

[UPDATE]
This specific doc (whether alone or together with other docs) returns empty string intermediate_step, which affects results:
https://github.com/vertinski/langchain_bugtest/blob/main/bug_doc-001.txt

[vertinski]

The issue was in prompting. I submitted a pull request ✌️
#660

[hwchase17]

thanks for raising and fixing!