Templates (#12294)

Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
Co-authored-by: Lance Martin <lance@langchain.dev>
Co-authored-by: Jacob Lee <jacoblee93@gmail.com>

templates/xml-agent/xml_agent/__init__.py

@@ -0,0 +1,3 @@
+from xml_agent.agent import agent_executor
+
+__all__ = ["agent_executor"]

templates/xml-agent/xml_agent/agent.py

@@ -0,0 +1,48 @@
+from langchain.chat_models import ChatAnthropic
+from langchain.tools.render import render_text_description
+from langchain.agents.format_scratchpad import format_xml
+from langchain.agents import AgentExecutor
+from langchain.retrievers.you import YouRetriever
+from langchain.agents.agent_toolkits.conversational_retrieval.tool import create_retriever_tool
+from langchain.pydantic_v1 import BaseModel
+from xml_agent.prompts import conversational_prompt, parse_output
+from langchain.schema import AIMessage, HumanMessage
+from typing import List, Tuple
+
+def _format_chat_history(chat_history: List[Tuple[str, str]]):
+    buffer = []
+    for human, ai in chat_history:
+        buffer.append(HumanMessage(content=human))
+        buffer.append(AIMessage(content=ai))
+    return buffer
+
+
+model = ChatAnthropic(model="claude-2")
+
+# Fake Tool
+retriever = YouRetriever(k=5)
+retriever_tool = create_retriever_tool(retriever, "search", "Use this to search for current events.")
+
+tools = [retriever_tool]
+
+prompt = conversational_prompt.partial(
+    tools=render_text_description(tools),
+    tool_names=", ".join([t.name for t in tools]),
+)
+llm_with_stop = model.bind(stop=["</tool_input>"])
+
+agent = {
+    "question": lambda x: x["question"],
+    "agent_scratchpad": lambda x: format_xml(x['intermediate_steps']),
+    "chat_history": lambda x: _format_chat_history(x["chat_history"]),
+} | prompt | llm_with_stop | parse_output
+
+class AgentInput(BaseModel):
+    question: str
+    chat_history: List[Tuple[str, str]]
+
+agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True, handle_parsing_errors=True).with_types(
+    input_type=AgentInput
+)
+
+agent_executor = agent_executor | (lambda x: x["output"])

templates/xml-agent/xml_agent/prompts.py

@@ -0,0 +1,50 @@
+from langchain.prompts import ChatPromptTemplate, MessagesPlaceholder
+from langchain.schema import AgentAction, AgentFinish
+
+template = """You are a helpful assistant. Help the user answer any questions.
+
+You have access to the following tools:
+
+{tools}
+
+In order to use a tool, you can use <tool></tool> and <tool_input></tool_input> tags. You will then get back a response in the form <observation></observation>
+For example, if you have a tool called 'search' that could run a google search, in order to search for the weather in SF you would respond:
+
+<tool>search</tool><tool_input>weather in SF</tool_input>
+<observation>64 degrees</observation>
+
+When you are done, you can respond as normal to the user.
+
+Example 1:
+
+Human: Hi!
+
+Assistant: Hi! How are you?
+
+Human: What is the weather in SF?
+Assistant: <tool>search</tool><tool_input>weather in SF</tool_input>
+<observation>64 degrees</observation>
+It is 64 degress in SF
+
+
+Begin!"""
+
+conversational_prompt = ChatPromptTemplate.from_messages([
+    ("system", template),
+    MessagesPlaceholder(variable_name="chat_history"),
+    ("user", "{question}"),
+    ("ai", "{agent_scratchpad}")
+])
+
+
+def parse_output(message):
+    text = message.content
+    if "</tool>" in text:
+        tool, tool_input = text.split("</tool>")
+        _tool = tool.split("<tool>")[1]
+        _tool_input = tool_input.split("<tool_input>")[1]
+        if "</tool_input>" in _tool_input:
+            _tool_input = _tool_input.split("</tool_input>")[0]
+        return AgentAction(tool=_tool, tool_input=_tool_input, log=text)
+    else:
+        return AgentFinish(return_values={"output": text}, log=text)