Hello did anybody else encounter error

HHayago

Hello, did anybody else encounter error with the guidance question generator for the subquestion query engine?

Example of error:
raise OutputParserException(
llama_index.output_parsers.base.OutputParserException: Got invalid JSON object. Error: Expecting property name enclosed in double quotes: line 2 column 14 (char 15) while parsing a flow mapping
in "<unicode string>", line 2, column 14:
"items": [{{#geneach 'items' stop=']'}}{{#u ...
^
expected ',' or '}', but got '<scalar>'
in "<unicode string>", line 3, column 48:
... ": "{{gen 'sub_question' stop='"'}}",

6 comments

LLogan M

interesting, I haven't seen this one 🤔 Seems like the JSON maybe got messed up because of quotes?

kkeychron

i am getting same

DDavid S

Hi I have the same issue using SubQuestionQueryEngine. But only if it needs to hit two indexes. If it uses one tool no problem. It use to work but I upgraded to 0.9.42

DDavid S

This is the error: raise OutputParserException(
llama_index.output_parsers.base.OutputParserException: Got invalid JSON object. Error: Expecting ',' delimiter: line 22 column 10 (char 670) while parsing a flow sequence
in "<unicode string>", line 2, column 14:
"items": [
^
expected ',' or ']', but got '<stream end>'
in "<unicode string>", line 22, column 10:

DDavid S

OK sorted. Increased max_new_tokens to 512. However did slow things down.

DDavid S

def createLLM(path: str, temp: float):
llm = LlamaCPP(
model_url=None,
model_path=path,
temperature=temp,
max_new_tokens=512,
context_window=20000,
generate_kwargs={},
model_kwargs={"n_gpu_layers": 4, "n_threads": 10, "n_threads_batch": 10},
# messages_to_prompt=messages_to_prompt,
# completion_to_prompt=completion_to_prompt,
verbose=True,
)
return llm

Add a reply

Find answers from the community

Hello did anybody else encounter error