`agent_memory` list index out of range

[casper-hansen]

Steps to reproduce

Launch server:

python -m mindsearch.app --lang en --model_format gpt4 --search_engine DuckDuckGoSearch --asy 

Launch streamlit:

streamlit run frontend/mindsearch_streamlit.py

Type in a query and get this error:

Traceback (most recent call last):
  File "/Users/cxrh/Documents/GitHub/MindSearch/mindsearch/app.py", line 142, in generate
    async for message in agent(inputs, session_id=session_id):
  File "/Users/cxrh/Documents/GitHub/MindSearch/mindsearch/agent/streaming.py", line 50, in __call__
    async for response_message in self.forward(*message, session_id=session_id, **kwargs):
  File "/Users/cxrh/Documents/GitHub/MindSearch/mindsearch/agent/mindsearch_agent.py", line 189, in forward
    reference, references_url = _generate_references_from_graph(gen.ret[1])
  File "/Users/cxrh/Documents/GitHub/MindSearch/mindsearch/agent/mindsearch_agent.py", line 41, in _generate_references_from_graph
    assert data_item["memory"]["agent.memory"][2]["sender"].endswith("ActionExecutor")
IndexError: list index out of range

[h66840]

i have the same problem

[casper-hansen]

@blankanswer @braisedpork1964 @lcolok I would really appreciate if you can try to reproduce and fix the error. I tried again today and there is no way around this error, it keeps happening on every query.

[nz3118Nan]

exactly same problem

[braisedpork1964]

Add examples to system prompts. See https://github.com/InternLM/MindSearch/issues/249#issuecomment-2472680114

I used BingSearch，for DDGS keeps reporting timeout errors.

• __init__.py

    agent = (AsyncMindSearchAgent if use_async else MindSearchAgent)(
        llm=llm,
        template=date,
        output_format=InterpreterParser(
            begin="<|action_start|><|interpreter|>",
            end="<|action_end|>",
            template=(
                GRAPH_PROMPT_CN + graph_fewshot_example_cn
                if lang == "cn"
                else GRAPH_PROMPT_EN + graph_fewshot_example_en
            ),
        ),
        searcher_cfg=dict(
            llm=llm,
            plugins=plugins,
            template=date,
            output_format=PluginParser(
                begin="<|action_start|><|plugin|>",
                end="<|action_end|>",
                template=(
                    searcher_system_prompt_cn + fewshot_example_cn
                    if lang == "cn"
                    else searcher_system_prompt_en + fewshot_example_en
                ),
                tool_info=get_plugin_prompt(plugins),
            ),
            user_input_template=(searcher_input_template_cn if lang == "cn" else searcher_input_template_en),
            user_context_template=(searcher_context_template_cn if lang == "cn" else searcher_context_template_en),
        ),
        summary_prompt=FINAL_RESPONSE_CN if lang == "cn" else FINAL_RESPONSE_EN,
        max_turn=10,
    )

• models.py

gpt4 = dict(
    type=GPTAPI,
    model_type="gpt-4-turbo",
    key=os.environ.get("OPENAI_API_KEY", "YOUR OPENAI API KEY"),
    api_base=os.environ.get("OPENAI_API_BASE", "https://api.openai.com/v1/chat/completions"),
    meta_template=[
        dict(role="system", api_role="system"),
        dict(role="user", api_role="user"),
        dict(role="assistant", api_role="assistant"),
        dict(role="environment", api_role="system"),
    ],
)

[casper-hansen]

Hi @braisedpork1964, I followed your instructions but I still get the same error after adding fewshot examples to InterpreterParser and PluginParser.

Is it possible to figure out a permanent fix for this and upstream it to the main branch?

[EmilyQian2001]

Hi @casper-hansen, did you solve this problem? I'm also using gpt4 and came across the same problem. Any help would be appreciated.

[casper-hansen]

@EmilyQian2001 No I didn't solve the problem. This bug is breaking the whole interaction, rendering MindSearch useless. I love the concept of MindSearch, but it's current state is that it does not work.

[EmilyQian2001]

@casper-hansen Same here. Did you successfully run it using internlm-2.5?