Spaces:
Paused
Paused
Create multiagent.py
Browse files- multiagent.py +336 -0
multiagent.py
ADDED
|
@@ -0,0 +1,336 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#Change to requirements caller
|
| 2 |
+
import sys
|
| 3 |
+
import subprocess
|
| 4 |
+
|
| 5 |
+
def run_pip_install():
|
| 6 |
+
packages = [
|
| 7 |
+
"langgraph",
|
| 8 |
+
"langchain",
|
| 9 |
+
"langchain_openai",
|
| 10 |
+
"langchain_experimental",
|
| 11 |
+
"qdrant-client",
|
| 12 |
+
"pymupdf",
|
| 13 |
+
"tiktoken",
|
| 14 |
+
"huggingface_hub",
|
| 15 |
+
"openai",
|
| 16 |
+
"tavily-python"
|
| 17 |
+
]
|
| 18 |
+
|
| 19 |
+
package_string = " ".join(packages)
|
| 20 |
+
|
| 21 |
+
try:
|
| 22 |
+
subprocess.check_call([sys.executable, "-m", "pip", "install", "-qU"] + packages)
|
| 23 |
+
print("All required packages have been installed successfully.")
|
| 24 |
+
except subprocess.CalledProcessError:
|
| 25 |
+
print(f"Failed to install packages. Please run the following command manually:")
|
| 26 |
+
print(f"%pip install -qU {package_string}")
|
| 27 |
+
sys.exit(1)
|
| 28 |
+
|
| 29 |
+
# Run pip install
|
| 30 |
+
run_pip_install()
|
| 31 |
+
|
| 32 |
+
import os
|
| 33 |
+
import functools
|
| 34 |
+
import operator
|
| 35 |
+
from typing import Annotated, List, Tuple, Union, Dict, Optional
|
| 36 |
+
from typing_extensions import TypedDict
|
| 37 |
+
import uuid
|
| 38 |
+
from pathlib import Path
|
| 39 |
+
|
| 40 |
+
from langchain_core.tools import tool
|
| 41 |
+
from langchain_core.messages import AIMessage, BaseMessage, HumanMessage
|
| 42 |
+
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
|
| 43 |
+
from langchain_openai import ChatOpenAI
|
| 44 |
+
from langchain.agents import AgentExecutor, create_openai_functions_agent
|
| 45 |
+
from langchain.output_parsers.openai_functions import JsonOutputFunctionsParser
|
| 46 |
+
from langchain_community.tools.tavily_search import TavilySearchResults
|
| 47 |
+
from langchain_community.vectorstores import Qdrant
|
| 48 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
| 49 |
+
from langchain_openai.embeddings import OpenAIEmbeddings
|
| 50 |
+
from langgraph.graph import END, StateGraph
|
| 51 |
+
from huggingface_hub import hf_hub_download
|
| 52 |
+
|
| 53 |
+
# Environment setup
|
| 54 |
+
OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY")
|
| 55 |
+
TAVILY_API_KEY = os.environ.get("TAVILY_API_KEY")
|
| 56 |
+
|
| 57 |
+
if not OPENAI_API_KEY:
|
| 58 |
+
raise ValueError("OPENAI_API_KEY not found in environment variables")
|
| 59 |
+
if not TAVILY_API_KEY:
|
| 60 |
+
raise ValueError("TAVILY_API_KEY not found in environment variables")
|
| 61 |
+
|
| 62 |
+
os.environ["OPENAI_API_KEY"] = OPENAI_API_KEY
|
| 63 |
+
os.environ["TAVILY_API_KEY"] = TAVILY_API_KEY
|
| 64 |
+
|
| 65 |
+
# CHANGE TO HF DIRECTORY
|
| 66 |
+
WORKING_DIRECTORY = Path("/tmp/content/data")
|
| 67 |
+
WORKING_DIRECTORY.mkdir(parents=True, exist_ok=True)
|
| 68 |
+
|
| 69 |
+
# Utility functions
|
| 70 |
+
def create_random_subdirectory():
|
| 71 |
+
random_id = str(uuid.uuid4())[:8]
|
| 72 |
+
subdirectory_path = WORKING_DIRECTORY / random_id
|
| 73 |
+
subdirectory_path.mkdir(exist_ok=True)
|
| 74 |
+
return subdirectory_path
|
| 75 |
+
|
| 76 |
+
def get_current_files():
|
| 77 |
+
try:
|
| 78 |
+
files = [f.relative_to(WORKING_DIRECTORY) for f in WORKING_DIRECTORY.rglob("*") if f.is_file()]
|
| 79 |
+
return "\n".join(str(f) for f in files) if files else "No files written."
|
| 80 |
+
except Exception:
|
| 81 |
+
return "Unable to retrieve current files."
|
| 82 |
+
|
| 83 |
+
# Document loading change to upload in HF
|
| 84 |
+
def fetch_hbr_article():
|
| 85 |
+
pdf_path = hf_hub_download(repo_id="your-username/your-repo-name", filename="murthy-loneliness.pdf")
|
| 86 |
+
return PyMuPDFLoader(pdf_path).load()
|
| 87 |
+
|
| 88 |
+
# Document processing
|
| 89 |
+
def tiktoken_len(text):
|
| 90 |
+
tokens = tiktoken.encoding_for_model("gpt-4o-mini").encode(text)
|
| 91 |
+
return len(tokens)
|
| 92 |
+
|
| 93 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
| 94 |
+
chunk_size=300,
|
| 95 |
+
chunk_overlap=0,
|
| 96 |
+
length_function=tiktoken_len,
|
| 97 |
+
)
|
| 98 |
+
|
| 99 |
+
docs = fetch_hbr_article()
|
| 100 |
+
split_chunks = text_splitter.split_documents(docs)
|
| 101 |
+
|
| 102 |
+
# Embedding and vector store setup
|
| 103 |
+
embedding_model = OpenAIEmbeddings(model="text-embedding-3-small")
|
| 104 |
+
qdrant_vectorstore = Qdrant.from_documents(
|
| 105 |
+
split_chunks,
|
| 106 |
+
embedding_model,
|
| 107 |
+
location=":memory:",
|
| 108 |
+
collection_name="extending_context_window_llama_3",
|
| 109 |
+
)
|
| 110 |
+
qdrant_retriever = qdrant_vectorstore.as_retriever()
|
| 111 |
+
|
| 112 |
+
# RAG setup
|
| 113 |
+
RAG_PROMPT = """
|
| 114 |
+
CONTEXT:
|
| 115 |
+
{context}
|
| 116 |
+
|
| 117 |
+
QUERY:
|
| 118 |
+
{question}
|
| 119 |
+
|
| 120 |
+
You are a helpful assistant. Use the available context to answer the question. If you can't answer the question, say you don't know.
|
| 121 |
+
"""
|
| 122 |
+
rag_prompt = ChatPromptTemplate.from_template(RAG_PROMPT)
|
| 123 |
+
openai_chat_model = ChatOpenAI(model="gpt-4o-mini")
|
| 124 |
+
|
| 125 |
+
rag_chain = (
|
| 126 |
+
{"context": itemgetter("question") | qdrant_retriever, "question": itemgetter("question")}
|
| 127 |
+
| rag_prompt | openai_chat_model | StrOutputParser()
|
| 128 |
+
)
|
| 129 |
+
|
| 130 |
+
# Tool definitions
|
| 131 |
+
@tool
|
| 132 |
+
def create_outline(points: List[str], file_name: str) -> str:
|
| 133 |
+
"""Create and save an outline."""
|
| 134 |
+
with (WORKING_DIRECTORY / file_name).open("w") as file:
|
| 135 |
+
for i, point in enumerate(points):
|
| 136 |
+
file.write(f"{i + 1}. {point}\n")
|
| 137 |
+
return f"Outline saved to {file_name}"
|
| 138 |
+
|
| 139 |
+
@tool
|
| 140 |
+
def read_document(file_name: str, start: Optional[int] = None, end: Optional[int] = None) -> str:
|
| 141 |
+
"""Read the specified document."""
|
| 142 |
+
with (WORKING_DIRECTORY / file_name).open("r") as file:
|
| 143 |
+
lines = file.readlines()
|
| 144 |
+
if start is not None:
|
| 145 |
+
start = 0
|
| 146 |
+
return "\n".join(lines[start:end])
|
| 147 |
+
|
| 148 |
+
@tool
|
| 149 |
+
def write_document(content: str, file_name: str) -> str:
|
| 150 |
+
"""Create and save a text document."""
|
| 151 |
+
with (WORKING_DIRECTORY / file_name).open("w") as file:
|
| 152 |
+
file.write(content)
|
| 153 |
+
return f"Document saved to {file_name}"
|
| 154 |
+
|
| 155 |
+
@tool
|
| 156 |
+
def edit_document(file_name: str, inserts: Dict[int, str] = {}) -> str:
|
| 157 |
+
"""Edit a document by inserting text at specific line numbers."""
|
| 158 |
+
with (WORKING_DIRECTORY / file_name).open("r") as file:
|
| 159 |
+
lines = file.readlines()
|
| 160 |
+
|
| 161 |
+
sorted_inserts = sorted(inserts.items())
|
| 162 |
+
for line_number, text in sorted_inserts:
|
| 163 |
+
if 1 <= line_number <= len(lines) + 1:
|
| 164 |
+
lines.insert(line_number - 1, text + "\n")
|
| 165 |
+
else:
|
| 166 |
+
return f"Error: Line number {line_number} is out of range."
|
| 167 |
+
|
| 168 |
+
with (WORKING_DIRECTORY / file_name).open("w") as file:
|
| 169 |
+
file.writelines(lines)
|
| 170 |
+
return f"Document edited and saved to {file_name}"
|
| 171 |
+
|
| 172 |
+
@tool
|
| 173 |
+
def retrieve_information(query: str):
|
| 174 |
+
"""Use Retrieval Augmented Generation to retrieve information about the 'murthy-loneliness' paper."""
|
| 175 |
+
return rag_chain.invoke({"question": query})
|
| 176 |
+
|
| 177 |
+
# Agent creation helpers
|
| 178 |
+
def create_team_agent(llm, tools, system_prompt, agent_name, team_members):
|
| 179 |
+
return create_agent(
|
| 180 |
+
llm,
|
| 181 |
+
tools,
|
| 182 |
+
f"{system_prompt}\nBelow are files currently in your directory:\n{{current_files}}",
|
| 183 |
+
team_members
|
| 184 |
+
)
|
| 185 |
+
|
| 186 |
+
def create_agent_node(agent, name):
|
| 187 |
+
return functools.partial(agent_node, agent=agent, name=name)
|
| 188 |
+
|
| 189 |
+
def add_agent_to_graph(graph, agent_name, agent_node):
|
| 190 |
+
graph.add_node(agent_name, agent_node)
|
| 191 |
+
graph.add_edge(agent_name, "supervisor")
|
| 192 |
+
|
| 193 |
+
def create_team_supervisor(llm, team_description, team_members):
|
| 194 |
+
return create_team_supervisor(
|
| 195 |
+
llm,
|
| 196 |
+
f"You are a supervisor tasked with managing a conversation between the"
|
| 197 |
+
f" following workers: {', '.join(team_members)}. {team_description}"
|
| 198 |
+
f" When all workers are finished, you must respond with FINISH.",
|
| 199 |
+
team_members
|
| 200 |
+
)
|
| 201 |
+
|
| 202 |
+
def create_team_chain(graph, team_members):
|
| 203 |
+
return (
|
| 204 |
+
functools.partial(enter_chain, members=team_members)
|
| 205 |
+
| graph.compile()
|
| 206 |
+
)
|
| 207 |
+
|
| 208 |
+
# LLM setup
|
| 209 |
+
llm = ChatOpenAI(model="gpt-4-turbo")
|
| 210 |
+
|
| 211 |
+
# Agent creation
|
| 212 |
+
tavily_tool = TavilySearchResults(max_results=5)
|
| 213 |
+
|
| 214 |
+
search_agent = create_team_agent(
|
| 215 |
+
llm,
|
| 216 |
+
[tavily_tool],
|
| 217 |
+
"You are a research assistant who can search for up-to-date info using the tavily search engine.",
|
| 218 |
+
"Search",
|
| 219 |
+
["Search", "PaperInformationRetriever"]
|
| 220 |
+
)
|
| 221 |
+
|
| 222 |
+
research_agent = create_team_agent(
|
| 223 |
+
llm,
|
| 224 |
+
[retrieve_information],
|
| 225 |
+
"You are a research assistant who can provide specific information on the provided paper: 'murthy-loneliness.pdf'. You must only respond with information about the paper related to the request.",
|
| 226 |
+
"PaperInformationRetriever",
|
| 227 |
+
["Search", "PaperInformationRetriever"]
|
| 228 |
+
)
|
| 229 |
+
|
| 230 |
+
doc_writer_agent = create_team_agent(
|
| 231 |
+
llm,
|
| 232 |
+
[write_document, edit_document, read_document],
|
| 233 |
+
"You are an expert writing technical social media posts.",
|
| 234 |
+
"DocWriter",
|
| 235 |
+
["DocWriter", "NoteTaker", "CopyEditor", "VoiceEditor"]
|
| 236 |
+
)
|
| 237 |
+
|
| 238 |
+
note_taking_agent = create_team_agent(
|
| 239 |
+
llm,
|
| 240 |
+
[create_outline, read_document],
|
| 241 |
+
"You are an expert senior researcher tasked with writing a social media post outline and taking notes to craft a social media post.",
|
| 242 |
+
"NoteTaker",
|
| 243 |
+
["DocWriter", "NoteTaker", "CopyEditor", "VoiceEditor"]
|
| 244 |
+
)
|
| 245 |
+
|
| 246 |
+
copy_editor_agent = create_team_agent(
|
| 247 |
+
llm,
|
| 248 |
+
[write_document, edit_document, read_document],
|
| 249 |
+
"You are an expert copy editor who focuses on fixing grammar, spelling, and tone issues.",
|
| 250 |
+
"CopyEditor",
|
| 251 |
+
["DocWriter", "NoteTaker", "CopyEditor", "VoiceEditor"]
|
| 252 |
+
)
|
| 253 |
+
|
| 254 |
+
voice_editor_agent = create_team_agent(
|
| 255 |
+
llm,
|
| 256 |
+
[write_document, edit_document, read_document],
|
| 257 |
+
"You are an expert in crafting and refining the voice and tone of social media posts. You edit the document to ensure it has a consistent, professional, and engaging voice appropriate for social media platforms.",
|
| 258 |
+
"VoiceEditor",
|
| 259 |
+
["DocWriter", "NoteTaker", "CopyEditor", "VoiceEditor"]
|
| 260 |
+
)
|
| 261 |
+
|
| 262 |
+
# Node creation
|
| 263 |
+
search_node = create_agent_node(search_agent, "Search")
|
| 264 |
+
research_node = create_agent_node(research_agent, "PaperInformationRetriever")
|
| 265 |
+
doc_writing_node = create_agent_node(doc_writer_agent, "DocWriter")
|
| 266 |
+
note_taking_node = create_agent_node(note_taking_agent, "NoteTaker")
|
| 267 |
+
copy_editing_node = create_agent_node(copy_editor_agent, "CopyEditor")
|
| 268 |
+
voice_node = create_agent_node(voice_editor_agent, "VoiceEditor")
|
| 269 |
+
|
| 270 |
+
# Graph creation
|
| 271 |
+
research_graph = StateGraph(ResearchTeamState)
|
| 272 |
+
add_agent_to_graph(research_graph, "Search", search_node)
|
| 273 |
+
add_agent_to_graph(research_graph, "PaperInformationRetriever", research_node)
|
| 274 |
+
|
| 275 |
+
authoring_graph = StateGraph(DocWritingState)
|
| 276 |
+
add_agent_to_graph(authoring_graph, "DocWriter", doc_writing_node)
|
| 277 |
+
add_agent_to_graph(authoring_graph, "NoteTaker", note_taking_node)
|
| 278 |
+
add_agent_to_graph(authoring_graph, "CopyEditor", copy_editing_node)
|
| 279 |
+
add_agent_to_graph(authoring_graph, "VoiceEditor", voice_node)
|
| 280 |
+
|
| 281 |
+
# Supervisor creation
|
| 282 |
+
research_supervisor = create_team_supervisor(
|
| 283 |
+
llm,
|
| 284 |
+
"Given the following user request, determine the subject to be researched and respond with the worker to act next.",
|
| 285 |
+
["Search", "PaperInformationRetriever"]
|
| 286 |
+
)
|
| 287 |
+
|
| 288 |
+
doc_writing_supervisor = create_team_supervisor(
|
| 289 |
+
llm,
|
| 290 |
+
"Given the following user request, determine which worker should act next. Each worker will perform a task and respond with their results and status.",
|
| 291 |
+
["DocWriter", "NoteTaker", "CopyEditor", "VoiceEditor"]
|
| 292 |
+
)
|
| 293 |
+
|
| 294 |
+
# Graph compilation
|
| 295 |
+
research_graph.add_node("supervisor", research_supervisor)
|
| 296 |
+
research_graph.set_entry_point("supervisor")
|
| 297 |
+
research_chain = create_team_chain(research_graph, research_graph.nodes)
|
| 298 |
+
|
| 299 |
+
authoring_graph.add_node("supervisor", doc_writing_supervisor)
|
| 300 |
+
authoring_graph.set_entry_point("supervisor")
|
| 301 |
+
authoring_chain = create_team_chain(authoring_graph, authoring_graph.nodes)
|
| 302 |
+
|
| 303 |
+
# Meta-supervisor setup
|
| 304 |
+
super_graph = StateGraph(State)
|
| 305 |
+
super_graph.add_node("Research team", get_last_message | research_chain | join_graph)
|
| 306 |
+
super_graph.add_node("SocialMedia team", get_last_message | authoring_chain | join_graph)
|
| 307 |
+
super_graph.add_node("supervisor", supervisor_node)
|
| 308 |
+
|
| 309 |
+
super_graph.add_edge("Research team", "supervisor")
|
| 310 |
+
super_graph.add_edge("SocialMedia team", "supervisor")
|
| 311 |
+
super_graph.add_conditional_edges(
|
| 312 |
+
"supervisor",
|
| 313 |
+
lambda x: x["next"],
|
| 314 |
+
{
|
| 315 |
+
"SocialMedia team": "SocialMedia team",
|
| 316 |
+
"Research team": "Research team",
|
| 317 |
+
"FINISH": END,
|
| 318 |
+
},
|
| 319 |
+
)
|
| 320 |
+
super_graph.set_entry_point("supervisor")
|
| 321 |
+
super_graph = super_graph.compile()
|
| 322 |
+
|
| 323 |
+
# Example usage
|
| 324 |
+
user_input = input("Enter your request for the social media post: ")
|
| 325 |
+
|
| 326 |
+
for s in super_graph.stream(
|
| 327 |
+
{
|
| 328 |
+
"messages": [
|
| 329 |
+
HumanMessage(content=user_input)
|
| 330 |
+
],
|
| 331 |
+
},
|
| 332 |
+
{"recursion_limit": 50},
|
| 333 |
+
):
|
| 334 |
+
if "__end__" not in s:
|
| 335 |
+
print(s)
|
| 336 |
+
print("---")
|