@ -674,153 +674,6 @@
"source": [
"agent.run(\"whats 2**.12\")"
]
},
{
"cell_type": "markdown",
"id": "8aa3c353-bd89-467c-9c27-b83a90cd4daa",
"metadata": {},
"source": [
"## Multi-argument tools\n",
"\n",
"Many functions expect structured inputs. These can also be supported using the Tool decorator or by directly subclassing `BaseTool`! We have to modify the LLM's OutputParser to map its string output to a dictionary to pass to the action, however."
]
},
{
"cell_type": "code",
"execution_count": 24,
"id": "537bc628",
"metadata": {
"tags": []
},
"outputs": [],
"source": [
"from typing import Optional, Union\n",
"\n",
"@tool\n",
"def custom_search(k: int, query: str, other_arg: Optional[str] = None):\n",
" \"\"\"The custom search function.\"\"\"\n",
" return f\"Here are the results for the custom search: k={k}, query={query}, other_arg={other_arg}\""
]
},
{
"cell_type": "code",
"execution_count": 25,
"id": "d5c992cf-776a-40cd-a6c4-e7cf65ea709e",
"metadata": {
"tags": []
},
"outputs": [],
"source": [
"import re\n",
"from langchain.schema import (\n",
" AgentAction,\n",
" AgentFinish,\n",
")\n",
"from langchain.agents import AgentOutputParser\n",
"\n",
"# We will add a custom parser to map the arguments to a dictionary\n",
"class CustomOutputParser(AgentOutputParser):\n",
" \n",
" def parse_tool_input(self, action_input: str) -> dict:\n",
" # Regex pattern to match arguments and their values\n",
" pattern = r\"(\\w+)\\s*=\\s*(None|\\\"[^\\\"]*\\\"|\\d+)\"\n",
" matches = re.findall(pattern, action_input)\n",
" \n",
" if not matches:\n",
" raise ValueError(f\"Could not parse action input: `{action_input}`\")\n",
"\n",
" # Create a dictionary with the parsed arguments and their values\n",
" parsed_input = {}\n",
" for arg, value in matches:\n",
" if value == \"None\":\n",
" parsed_value = None\n",
" elif value.isdigit():\n",
" parsed_value = int(value)\n",
" else:\n",
" parsed_value = value.strip('\"')\n",
" parsed_input[arg] = parsed_value\n",
"\n",
" return parsed_input\n",
" \n",
" def parse(self, llm_output: str) -> Union[AgentAction, AgentFinish]:\n",
" # Check if agent should finish\n",
" if \"Final Answer:\" in llm_output:\n",
" return AgentFinish(\n",
" # Return values is generally always a dictionary with a single `output` key\n",
" # It is not recommended to try anything else at the moment :)\n",
" return_values={\"output\": llm_output.split(\"Final Answer:\")[-1].strip()},\n",
" log=llm_output,\n",
" )\n",
" # Parse out the action and action input\n",
" regex = r\"Action\\s*\\d*\\s*:(.*?)\\nAction\\s*\\d*\\s*Input\\s*\\d*\\s*:[\\s]*(.*)\"\n",
" match = re.search(regex, llm_output, re.DOTALL)\n",
" if not match:\n",
" raise ValueError(f\"Could not parse LLM output: `{llm_output}`\")\n",
" action = match.group(1).strip()\n",
" action_input = match.group(2)\n",
" tool_input = self.parse_tool_input(action_input)\n",
" # Return the action and action \n",
" return AgentAction(tool=action, tool_input=tool_input, log=llm_output)"
]
},
{
"cell_type": "code",
"execution_count": 26,
"id": "68269547-1482-4138-a6ea-58f00b4a9548",
"metadata": {
"tags": []
},
"outputs": [],
"source": [
"llm = OpenAI(temperature=0)\n",
"agent = initialize_agent([custom_search], llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True, agent_kwargs={\"output_parser\": CustomOutputParser()})"
]
},
{
"cell_type": "code",
"execution_count": 27,
"id": "0947835a-691c-4f51-b8f4-6744e0e48ab1",
"metadata": {
"tags": []
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n",
"\n",
"\u001b[1m> Entering new AgentExecutor chain...\u001b[0m\n",
"\u001b[32;1m\u001b[1;3m I need to use a search function to find the answer\n",
"Action: custom_search\n",
"Action Input: k=1, query=\"me\"\u001b[0m\u001b[36;1m\u001b[1;3mHere are the results for the custom search: k=1, query=me, other_arg=None\u001b[0m\u001b[32;1m\u001b[1;3m I now know the final answer\n",
"Final Answer: The results of the custom search for k=1, query=me, other_arg=None.\u001b[0m\n",
"\n",
"\u001b[1m> Finished chain.\u001b[0m\n"
]
},
{
"data": {
"text/plain": [
"'The results of the custom search for k=1, query=me, other_arg=None.'"
]
},
"execution_count": 27,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"agent.run(\"Search for me and tell me whatever it says\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "caf39c66-102b-42c1-baf2-777a49886ce4",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
@ -839,7 +692,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.1 "
"version": "3.11.2 "
},
"vscode": {
"interpreter": {
Zander Chase
1 year ago
committed by
GitHub