Spaces:
Sleeping
Sleeping
Update rag_langgraph.py
Browse files- rag_langgraph.py +84 -109
rag_langgraph.py
CHANGED
|
@@ -1,12 +1,20 @@
|
|
| 1 |
-
import
|
| 2 |
-
|
| 3 |
from langchain_core.messages import (
|
|
|
|
| 4 |
BaseMessage,
|
| 5 |
ToolMessage,
|
| 6 |
HumanMessage,
|
| 7 |
)
|
| 8 |
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
from langgraph.graph import END, StateGraph
|
|
|
|
|
|
|
|
|
|
| 10 |
|
| 11 |
def create_agent(llm, tools, system_message: str):
|
| 12 |
"""Create an agent."""
|
|
@@ -29,21 +37,8 @@ def create_agent(llm, tools, system_message: str):
|
|
| 29 |
prompt = prompt.partial(tool_names=", ".join([tool.name for tool in tools]))
|
| 30 |
return prompt | llm.bind_tools(tools)
|
| 31 |
|
| 32 |
-
from langchain_core.tools import tool
|
| 33 |
-
from typing import Annotated
|
| 34 |
-
from langchain_experimental.utilities import PythonREPL
|
| 35 |
-
from langchain_community.tools.tavily_search import TavilySearchResults
|
| 36 |
-
|
| 37 |
-
tavily_tool = TavilySearchResults(max_results=5)
|
| 38 |
-
|
| 39 |
-
# Warning: This executes code locally, which can be unsafe when not sandboxed
|
| 40 |
-
|
| 41 |
-
repl = PythonREPL()
|
| 42 |
-
|
| 43 |
@tool
|
| 44 |
-
def python_repl(
|
| 45 |
-
code: Annotated[str, "The python code to execute to generate your chart."]
|
| 46 |
-
):
|
| 47 |
"""Use this to execute python code. If you want to see the output of a value,
|
| 48 |
you should print it out with `print(...)`. This is visible to the user."""
|
| 49 |
try:
|
|
@@ -55,23 +50,12 @@ def python_repl(
|
|
| 55 |
result_str + "\n\nIf you have completed all tasks, respond with FINAL ANSWER."
|
| 56 |
)
|
| 57 |
|
| 58 |
-
import operator
|
| 59 |
-
from typing import Annotated, Sequence, TypedDict
|
| 60 |
-
|
| 61 |
-
from langchain_openai import ChatOpenAI
|
| 62 |
-
from typing_extensions import TypedDict
|
| 63 |
-
|
| 64 |
-
|
| 65 |
# This defines the object that is passed between each node
|
| 66 |
# in the graph. We will create different nodes for each agent and tool
|
| 67 |
class AgentState(TypedDict):
|
| 68 |
messages: Annotated[Sequence[BaseMessage], operator.add]
|
| 69 |
sender: str
|
| 70 |
|
| 71 |
-
import functools
|
| 72 |
-
from langchain_core.messages import AIMessage
|
| 73 |
-
|
| 74 |
-
|
| 75 |
# Helper function to create a node for a given agent
|
| 76 |
def agent_node(state, agent, name):
|
| 77 |
result = agent.invoke(state)
|
|
@@ -87,32 +71,6 @@ def agent_node(state, agent, name):
|
|
| 87 |
"sender": name,
|
| 88 |
}
|
| 89 |
|
| 90 |
-
llm = ChatOpenAI(model="gpt-4-1106-preview")
|
| 91 |
-
|
| 92 |
-
# Research agent and node
|
| 93 |
-
research_agent = create_agent(
|
| 94 |
-
llm,
|
| 95 |
-
[tavily_tool],
|
| 96 |
-
system_message="You should provide accurate data for the chart_generator to use.",
|
| 97 |
-
)
|
| 98 |
-
research_node = functools.partial(agent_node, agent=research_agent, name="Researcher")
|
| 99 |
-
|
| 100 |
-
# chart_generator
|
| 101 |
-
chart_agent = create_agent(
|
| 102 |
-
llm,
|
| 103 |
-
[python_repl],
|
| 104 |
-
system_message="Any charts you display will be visible by the user.",
|
| 105 |
-
)
|
| 106 |
-
chart_node = functools.partial(agent_node, agent=chart_agent, name="chart_generator")
|
| 107 |
-
|
| 108 |
-
from langgraph.prebuilt import ToolNode
|
| 109 |
-
|
| 110 |
-
tools = [tavily_tool, python_repl]
|
| 111 |
-
tool_node = ToolNode(tools)
|
| 112 |
-
|
| 113 |
-
# Either agent can decide to end
|
| 114 |
-
from typing import Literal
|
| 115 |
-
|
| 116 |
def router(state) -> Literal["call_tool", "__end__", "continue"]:
|
| 117 |
# This is the router
|
| 118 |
messages = state["messages"]
|
|
@@ -125,62 +83,79 @@ def router(state) -> Literal["call_tool", "__end__", "continue"]:
|
|
| 125 |
return "__end__"
|
| 126 |
return "continue"
|
| 127 |
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
workflow.add_node("chart_generator", chart_node)
|
| 132 |
-
workflow.add_node("call_tool", tool_node)
|
| 133 |
-
|
| 134 |
-
workflow.add_conditional_edges(
|
| 135 |
-
"Researcher",
|
| 136 |
-
router,
|
| 137 |
-
{"continue": "chart_generator", "call_tool": "call_tool", "__end__": END},
|
| 138 |
-
)
|
| 139 |
-
workflow.add_conditional_edges(
|
| 140 |
-
"chart_generator",
|
| 141 |
-
router,
|
| 142 |
-
{"continue": "Researcher", "call_tool": "call_tool", "__end__": END},
|
| 143 |
-
)
|
| 144 |
-
|
| 145 |
-
workflow.add_conditional_edges(
|
| 146 |
-
"call_tool",
|
| 147 |
-
# Each agent node updates the 'sender' field
|
| 148 |
-
# the tool calling node does not, meaning
|
| 149 |
-
# this edge will route back to the original agent
|
| 150 |
-
# who invoked the tool
|
| 151 |
-
lambda x: x["sender"],
|
| 152 |
-
{
|
| 153 |
-
"Researcher": "Researcher",
|
| 154 |
-
"chart_generator": "chart_generator",
|
| 155 |
-
},
|
| 156 |
-
)
|
| 157 |
-
workflow.set_entry_point("Researcher")
|
| 158 |
-
graph = workflow.compile()
|
| 159 |
-
|
| 160 |
-
from IPython.display import Image, display
|
| 161 |
|
| 162 |
-
|
| 163 |
-
display(Image(graph.get_graph(xray=True).draw_mermaid_png()))
|
| 164 |
-
except:
|
| 165 |
-
# This requires some extra dependencies and is optional
|
| 166 |
-
pass
|
| 167 |
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
|
| 172 |
-
|
| 173 |
-
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
|
| 181 |
-
|
| 182 |
-
|
| 183 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 184 |
|
| 185 |
-
|
| 186 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import functools, operator
|
| 2 |
+
from IPython.display import Image, display
|
| 3 |
from langchain_core.messages import (
|
| 4 |
+
AIMessage,
|
| 5 |
BaseMessage,
|
| 6 |
ToolMessage,
|
| 7 |
HumanMessage,
|
| 8 |
)
|
| 9 |
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
|
| 10 |
+
from langchain_core.tools import tool
|
| 11 |
+
from langchain_community.tools.tavily_search import TavilySearchResults
|
| 12 |
+
from langchain_experimental.utilities import PythonREPL
|
| 13 |
+
from langchain_openai import ChatOpenAI
|
| 14 |
from langgraph.graph import END, StateGraph
|
| 15 |
+
from langgraph.prebuilt import ToolNode
|
| 16 |
+
from typing import Annotated, Literal, Sequence, TypedDict
|
| 17 |
+
from typing_extensions import TypedDict
|
| 18 |
|
| 19 |
def create_agent(llm, tools, system_message: str):
|
| 20 |
"""Create an agent."""
|
|
|
|
| 37 |
prompt = prompt.partial(tool_names=", ".join([tool.name for tool in tools]))
|
| 38 |
return prompt | llm.bind_tools(tools)
|
| 39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
@tool
|
| 41 |
+
def python_repl(code: Annotated[str, "The python code to execute to generate your chart."]):
|
|
|
|
|
|
|
| 42 |
"""Use this to execute python code. If you want to see the output of a value,
|
| 43 |
you should print it out with `print(...)`. This is visible to the user."""
|
| 44 |
try:
|
|
|
|
| 50 |
result_str + "\n\nIf you have completed all tasks, respond with FINAL ANSWER."
|
| 51 |
)
|
| 52 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
# This defines the object that is passed between each node
|
| 54 |
# in the graph. We will create different nodes for each agent and tool
|
| 55 |
class AgentState(TypedDict):
|
| 56 |
messages: Annotated[Sequence[BaseMessage], operator.add]
|
| 57 |
sender: str
|
| 58 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
# Helper function to create a node for a given agent
|
| 60 |
def agent_node(state, agent, name):
|
| 61 |
result = agent.invoke(state)
|
|
|
|
| 71 |
"sender": name,
|
| 72 |
}
|
| 73 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
def router(state) -> Literal["call_tool", "__end__", "continue"]:
|
| 75 |
# This is the router
|
| 76 |
messages = state["messages"]
|
|
|
|
| 83 |
return "__end__"
|
| 84 |
return "continue"
|
| 85 |
|
| 86 |
+
def run_multi_agent(prompt):
|
| 87 |
+
tavily_tool = TavilySearchResults(max_results=5)
|
| 88 |
+
repl = PythonREPL()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 89 |
|
| 90 |
+
llm = ChatOpenAI(model="gpt-4o")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 91 |
|
| 92 |
+
# Research agent and node
|
| 93 |
+
research_agent = create_agent(
|
| 94 |
+
llm,
|
| 95 |
+
[tavily_tool],
|
| 96 |
+
system_message="You should provide accurate data for the chart_generator to use.",
|
| 97 |
+
)
|
| 98 |
+
research_node = functools.partial(agent_node, agent=research_agent, name="Researcher")
|
| 99 |
+
|
| 100 |
+
# chart_generator
|
| 101 |
+
chart_agent = create_agent(
|
| 102 |
+
llm,
|
| 103 |
+
[python_repl],
|
| 104 |
+
system_message="Any charts you display will be visible by the user.",
|
| 105 |
+
)
|
| 106 |
+
chart_node = functools.partial(agent_node, agent=chart_agent, name="chart_generator")
|
| 107 |
+
|
| 108 |
+
tools = [tavily_tool, python_repl]
|
| 109 |
+
tool_node = ToolNode(tools)
|
| 110 |
+
|
| 111 |
+
workflow = StateGraph(AgentState)
|
| 112 |
+
|
| 113 |
+
workflow.add_node("Researcher", research_node)
|
| 114 |
+
workflow.add_node("chart_generator", chart_node)
|
| 115 |
+
workflow.add_node("call_tool", tool_node)
|
| 116 |
+
|
| 117 |
+
workflow.add_conditional_edges(
|
| 118 |
+
"Researcher",
|
| 119 |
+
router,
|
| 120 |
+
{"continue": "chart_generator", "call_tool": "call_tool", "__end__": END},
|
| 121 |
+
)
|
| 122 |
+
workflow.add_conditional_edges(
|
| 123 |
+
"chart_generator",
|
| 124 |
+
router,
|
| 125 |
+
{"continue": "Researcher", "call_tool": "call_tool", "__end__": END},
|
| 126 |
+
)
|
| 127 |
+
|
| 128 |
+
workflow.add_conditional_edges(
|
| 129 |
+
"call_tool",
|
| 130 |
+
# Each agent node updates the 'sender' field
|
| 131 |
+
# the tool calling node does not, meaning
|
| 132 |
+
# this edge will route back to the original agent
|
| 133 |
+
# who invoked the tool
|
| 134 |
+
lambda x: x["sender"],
|
| 135 |
+
{
|
| 136 |
+
"Researcher": "Researcher",
|
| 137 |
+
"chart_generator": "chart_generator",
|
| 138 |
+
},
|
| 139 |
+
)
|
| 140 |
+
workflow.set_entry_point("Researcher")
|
| 141 |
+
graph = workflow.compile()
|
| 142 |
|
| 143 |
+
try:
|
| 144 |
+
display(Image(graph.get_graph(xray=True).draw_mermaid_png()))
|
| 145 |
+
except:
|
| 146 |
+
# This requires some extra dependencies and is optional
|
| 147 |
+
pass
|
| 148 |
+
|
| 149 |
+
events = graph.stream(
|
| 150 |
+
{
|
| 151 |
+
"messages": [
|
| 152 |
+
HumanMessage(
|
| 153 |
+
content=prompt
|
| 154 |
+
)
|
| 155 |
+
],
|
| 156 |
+
},
|
| 157 |
+
# Maximum number of steps to take in the graph
|
| 158 |
+
{"recursion_limit": 150},
|
| 159 |
+
)
|
| 160 |
+
for s in events:
|
| 161 |
+
return s
|