FunctionTool, the output appears to be hallucinated. Thought: The user has provided his preference. I can use the select_game tool to find the appropriate game for him.
Action: select_game
Action Input: {"preference_description": "racing"}
Observation: {"game_name": "Forza Horizon"}select_game will never give the above ToolOutput.Functiontool.fn as it's Observation?tools = []
def select_game(preference_description: str) -> Dict[str, str]:
"""Use the user's description of his preference to select a game for him."""
return random.choice([{
"game_name": "Final Fantasy",
"webpage": "http://ff7.game",
}, {
"game_name": "Call of Duty",
"webpage": "http://cod.game",
}])
tools.append(FunctionTool.from_defaults(fn=select_game))
agent = ReActAgent.from_tools(tools, llm=llm, verbose=True)Action Input and Observation, so maybe this causes the react steps to be part of the agent response...sql_database = SQLDatabase(engine, include_tables=table_names)
sql_query_engine = NLSQLTableQueryEngine(sql_database=sql_database)
sql_tool = QueryEngineTool.from_defaults(
query_engine=sql_query_engine,
description=(
"Useful for translating a natural language query into an SQL query over tables containing:"
"exam_scores, containing information about student's grades in an exam. "
# and descriptions of more tables...
),
)def alpha(num_unique_subjects: int) -> int:
return num_unique_subjects * num_unique_subjects
alpha_tool = FunctionTool.from_defaults(fn=alpha)query_engine_tools = [sql_tool, alpha_tool] agent = OpenAIAgent.from_tools(query_engine_tools, verbose=True) agent.chat_repl()
alpha formula passing in a hallucinated value for num_unique_subjects .from llama_index.core.prompts import PromptTemplate
from llama_index.core.prompts.system import SHAKESPEARE_WRITING_ASSISTANT
from llama_index.core.agent import ReActAgent
SHAKESPEARE_WRITING_ASSISTANT = PromptTemplate(SHAKESPEARE_WRITING_ASSISTANT) # just a random example
agent = ReActAgent.from_tools(
[],
llm=llm,
verbose=True,
)
agent.update_prompts({"agent_worker:system_prompt": SHAKESPEARE_WRITING_ASSISTANT}) File "/home/x/anaconda3/envs/foo/lib/python3.10/site-packages/llama_index/core/prompts/mixin.py", line 75, in update_prompts
raise ValueError(f"Module {module_name} not found.")
ValueError: Module agent_worker not found.FunctionTools named foo_tool and bar_tool that I want to use in a QueryPlanTool which only accepts QueryEngineTools.def foo():
return "foo"
def bar():
return "bar"
foo_tool = FunctionTool.from_defaults(fn=foo)
bar_tool = FunctionTool.from_defaults(fn=bar)FunctionTool to QueryEngineTool? Or create QueryEngineTool with the same logic as foo and bar?