AttributeError: 'list' object has no attribute 'split' when running deep research

[TineKolenik]

When I try to use GPTR's deep research (PIP package, Version: 0.12.12; python 3.13), I get

File "..\gpt_researcher\skills\deep_research.py", line 17, in count_words return len(text.split())

This happens after already collecting, verifying and ranking top sources.

This is my config.json:

{
    "RETRIEVER": "duckduckgo, arxiv",
    "EMBEDDING": "openai:text-embedding-3-small",
    "FAST_LLM": "openai:gpt-4o-mini",
    "SMART_LLM": "openai:gpt-4o",
    "STRATEGIC_LLM": "openai:gpt-4o",
    "LANGUAGE": "english",
    "CURATE_SOURCES": true,
    "FAST_TOKEN_LIMIT": 2000,
    "SMART_TOKEN_LIMIT": 4000,
    "STRATEGIC_TOKEN_LIMIT": 4000,
    "BROWSE_CHUNK_MAX_LENGTH": 8192,
    "SUMMARY_TOKEN_LIMIT": 700,
    "TEMPERATURE": 0.55,
    "REPORT_FORMAT": "IEEE",
    "MAX_ITERATIONS": 3,
    "AGENT_ROLE": null,
    "MAX_SUBTOPICS": 5,
    "SCRAPER": "bs",
    "MAX_SCRAPER_WORKERS": 15,
    "DOC_PATH": null,
    "USER_AGENT": null,
    "MEMORY_BACKEND": "local",
    "deep_research_breadth": 4,
    "deep_research_depth": 2,
    "deep_research_concurrency": 4,
    "total_words": 2500
}

My code:

from gpt_researcher import GPTResearcher
from gpt_researcher.utils.enum import ReportType, Tone
import asyncio
import os

os.environ["OPENAI_API_KEY"] = MY_KEY

query = MY_QUERY

async def main():
    # Initialize researcher with deep research type
    researcher = GPTResearcher(
        query=query,
        report_type="deep",
        config_path="config_research_deep.json"
    )

    # Run research
    research_data = await researcher.conduct_research()

    # Generate report
    report = await researcher.write_report()
    print(report)


if __name__ == "__main__":
    asyncio.run(main())

I also tried a yaml, but that didn't work, despite https://docs.gptr.dev/docs/gpt-researcher/gptr/deep_research mentioning a yaml file.

BTW, normal research works fine.

[ElishaKay]

Welcome @TineKolenik

Can you try to minimize the values within your config.json and try again?

Alternatively, remove the config_path parameter from the GPTResearcher run - instead, create a minimalistic .env file in your root directory with:

OPENAI_API_KEY=
TAVILY_API_KEY=
DOC_PATH=./my-docs

Then, try running your script again and let us know if it fails - I'd like to try to get to the root cause

[TineKolenik]

@ElishaKay thanks for the welcome and the quick help. Here are my findings:

Suggestions 1 - Config.json value minimization:

I reduced the config.json to:

{
    "EMBEDDING": "openai:text-embedding-3-small",
    "FAST_LLM": "openai:gpt-4o-mini",
    "SMART_LLM": "openai:gpt-4o",
    "STRATEGIC_LLM": "openai:gpt-4o",
    "LANGUAGE": "english",
    "CURATE_SOURCES": true,
    "deep_research_breadth": 4,
    "deep_research_depth": 2,
    "deep_research_concurrency": 4,
    "total_words": 2500
}

And I added a Tavily API key to use for web search. This reproduced the same error. As before, it does a bunch of research, calculates costs, ranks sources, etc., but in the end comes up with the same error.

Suggestion 2 - config_path removal and .env utilization:

This worked successfully, however, the result was very lacking and short (below 800 words, even though the default length is 2500). The culprit does seem to be either the config_path or one of the values still present above, and I feel that having a config is absolutely necessary to generate good results. Do you have any idea how to produce longer lengths without utilizing config or how to get it to work with config_path? This would also be useful for other stuff (citation style, if I wanted to change to another web search option ...).

Thanks!