name-that-trait / app.py
Dirk Haupt
Revert to after adding supervisord
6f120f4
import chainlit as cl
from langchain_anthropic import ChatAnthropic
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain_qdrant import QdrantVectorStore
from qdrant_client import QdrantClient
from qdrant_client.http.models import Distance, VectorParams
from langchain.prompts import ChatPromptTemplate
from langchain.text_splitter import RecursiveCharacterTextSplitter
from pathlib import Path
import os
from huggingface_hub import HfApi, hf_hub_download
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from typing import Optional
import asyncio
from contextlib import asynccontextmanager
# Global variable to store the vectorstore instance
_vectorstore = None
# Configurable model selection
MODELS = {
"gpt-3.5-turbo": ChatOpenAI(model_name="gpt-3.5-turbo"),
"gpt-4": ChatOpenAI(model_name="gpt-4"),
"claude-3-sonnet": ChatAnthropic(model="claude-3-sonnet")
}
# Custom prompt for vegan ethics
VEGAN_ETHICS_TEMPLATE = """
You are an ethical philosopher specializing in animal rights and veganism.
Use the following pieces of context to answer the user's question.
Always guide your response toward applying the "Name the Trait" argument:
1. Ask what trait in non-human animals, if present in humans, would justify treating humans the same way
2. Examine if this trait is morally relevant or arbitrary
3. Challenge the user to maintain logical consistency in their ethical framework
4. Use scientific evidence from the context to support your points
Context: {context}
Question: {question}
Answer:
"""
def get_vectorstore(persist_dir: str = "vector_store"):
"""Create or return cached vectorstore instance"""
global _vectorstore
if _vectorstore is not None:
return _vectorstore
# Initialize vector store with persistence
persist_dir = Path(persist_dir)
client = QdrantClient(
path=str(persist_dir),
force_disable_check_same_thread=True # Important for concurrent access
)
# Check if collection exists
collections = client.get_collections().collections
collection_names = [c.name for c in collections]
if "vegan_ethics" not in collection_names:
print(f"Creating new vector store in {persist_dir}")
client.create_collection(
collection_name="vegan_ethics",
vectors_config=VectorParams(
size=1536,
distance=Distance.COSINE,
),
)
_vectorstore = QdrantVectorStore(
client=client,
embedding=OpenAIEmbeddings(),
collection_name="vegan_ethics"
)
return _vectorstore
async def process_and_load_documents(vectorstore, repo_id="Frikster42/name-that-trait", data_folder="data"):
# Create a single TaskList for the entire process
tasks = cl.TaskList()
tasks.status = "Initializing..."
await tasks.send()
msg = cl.Message(content="Loading documents from Hugging Face repository... please be patient...")
await msg.send()
data_dir = Path("data")
data_dir.mkdir(exist_ok=True)
# Get list of files in the repository
api = HfApi()
dataset_files = api.list_repo_files(repo_id, repo_type="dataset")
dataset_pdf_files = [f for f in dataset_files if f.endswith('.pdf')]
# Download phase
tasks.status = "Downloading files..."
await tasks.send()
for i, pdf_file in enumerate(dataset_pdf_files):
task = cl.Task(title=f"Downloading {pdf_file}")
await tasks.add_task(task)
hf_hub_download(
repo_id=repo_id,
filename=pdf_file,
local_dir=str(data_dir),
local_dir_use_symlinks=False,
repo_type="dataset"
)
task.status = cl.TaskStatus.DONE
await tasks.send()
# Loading phase
documents = []
pdf_files = [f for f in os.listdir(data_folder) if f.endswith('.pdf')]
tasks.status = "Loading files..."
await tasks.send()
for i, filename in enumerate(pdf_files):
task = cl.Task(title=f"Loading {filename}")
await tasks.add_task(task)
filepath = os.path.join(data_folder, filename)
if filename.endswith('.pdf'):
from langchain.document_loaders import PyPDFLoader
loader = PyPDFLoader(filepath)
else:
from langchain.document_loaders import TextLoader
loader = TextLoader(filepath)
documents.extend(loader.load())
task.status = cl.TaskStatus.DONE
await tasks.send()
# Split and process documents
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
chunks = text_splitter.split_documents(documents)
if chunks:
tasks.status = "Processing chunks..."
await tasks.send()
batch_size = 100
num_batches = (len(chunks) + batch_size - 1) // batch_size
for i in range(0, len(chunks), batch_size):
task = cl.Task(title=f"Processing batch {(i//batch_size)+1}/{num_batches}")
await tasks.add_task(task)
batch = chunks[i:i + batch_size]
vectorstore.add_documents(batch)
task.status = cl.TaskStatus.DONE
await tasks.send()
tasks.status = "Completed"
await tasks.send()
msg = cl.Message(content="βœ… Documents loaded successfully!")
await msg.send()
return vectorstore
# Create FastAPI app
app = FastAPI(title="Vegan Ethics RAG API")
class QueryRequest(BaseModel):
question: str
model_name: Optional[str] = "gpt-3.5-turbo"
@app.post("/api/query")
async def query_endpoint(request: QueryRequest):
try:
# Get or create vectorstore instance
vectorstore = get_vectorstore()
# Create prompt template
prompt = ChatPromptTemplate.from_messages([
("system", VEGAN_ETHICS_TEMPLATE),
("user", "{question}")
])
# Validate model selection
if request.model_name not in MODELS:
raise HTTPException(status_code=400, detail=f"Invalid model name. Choose from: {list(MODELS.keys())}")
# Get relevant documents
docs = vectorstore.similarity_search(request.question, k=3)
context = "\n".join(doc.page_content for doc in docs)
# Generate response
chain = prompt | MODELS[request.model_name]
response = await chain.ainvoke({
"context": context,
"question": request.question
})
return {
"response": response.content,
"model_used": request.model_name
}
except Exception as e:
raise HTTPException(status_code=500, detail=str(e))
@cl.on_chat_start
async def start():
# Get or create vectorstore instance
vectorstore = get_vectorstore()
# Load documents if needed
collection_info = vectorstore.client.get_collection("vegan_ethics")
if collection_info.points_count == 0:
await cl.Message(content="Vector store is empty, loading documents...").send()
vectorstore = await process_and_load_documents(vectorstore)
# Create prompt template
prompt = ChatPromptTemplate.from_messages([
("system", VEGAN_ETHICS_TEMPLATE),
("user", "{question}")
])
# Store components in session
cl.user_session.set("vectorstore", vectorstore)
cl.user_session.set("prompt", prompt)
cl.user_session.set("model_name", "gpt-3.5-turbo")
# UI for model selection
actions = [
cl.Action(
name="model_select",
label="Current Model: gpt-3.5-turbo",
description="Change the AI model",
payload={"current_model": "gpt-3.5-turbo"}
)
]
await cl.Message(
content="Welcome to the Vegan Ethics Assistant. Ask any question about veganism, ethics, or animal consumption.",
actions=actions
).send()
@cl.action_callback("model_select")
async def on_action(action):
models_list = list(MODELS.keys())
current_index = models_list.index(action.payload["current_model"])
next_index = (current_index + 1) % len(models_list)
next_model_name = models_list[next_index]
cl.user_session.set("model_name", next_model_name)
actions = [
cl.Action(
name="model_select",
label=f"Current Model: {next_model_name}",
description="Change the AI model",
payload={"current_model": next_model_name}
)
]
await cl.Message(content=f"Model switched to {next_model_name}", actions=actions).send()
@cl.on_message
async def main(message):
vectorstore = cl.user_session.get("vectorstore")
prompt = cl.user_session.get("prompt")
model_name = cl.user_session.get("model_name")
# Get relevant documents
docs = vectorstore.similarity_search(message.content, k=3)
context = "\n".join(doc.page_content for doc in docs)
# Generate response
chain = prompt | MODELS[model_name]
response = await chain.ainvoke({
"context": context,
"question": message.content
})
await cl.Message(content=response.content).send()