Spaces:
Sleeping
Sleeping
Create main.py
Browse files
main.py
ADDED
@@ -0,0 +1,111 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from fastapi import FastAPI, HTTPException
|
2 |
+
from pydantic import BaseModel
|
3 |
+
from langchain.chains import RetrievalQA
|
4 |
+
from langchain.prompts import PromptTemplate
|
5 |
+
from langchain_community.vectorstores import FAISS
|
6 |
+
from langchain_huggingface import HuggingFaceEmbeddings
|
7 |
+
from langchain_groq import ChatGroq
|
8 |
+
import zipfile
|
9 |
+
import os
|
10 |
+
|
11 |
+
app = FastAPI()
|
12 |
+
|
13 |
+
# === Startup config ===
|
14 |
+
class QueryRequest(BaseModel):
|
15 |
+
question: str
|
16 |
+
|
17 |
+
llm = None
|
18 |
+
retriever = None
|
19 |
+
chain = None
|
20 |
+
|
21 |
+
@app.on_event("startup")
|
22 |
+
def load_components():
|
23 |
+
global llm, retriever, chain
|
24 |
+
|
25 |
+
api_key = os.getenv('api_key')
|
26 |
+
|
27 |
+
# --- Load LLM ---
|
28 |
+
llm = ChatGroq(
|
29 |
+
model="meta-llama/llama-4-scout-17b-16e-instruct",
|
30 |
+
temperature=0,
|
31 |
+
max_tokens=1024,
|
32 |
+
api_key=api_key
|
33 |
+
)
|
34 |
+
|
35 |
+
# --- Load Embeddings ---
|
36 |
+
embeddings = HuggingFaceEmbeddings(
|
37 |
+
model_name="intfloat/multilingual-e5-large",
|
38 |
+
model_kwargs={"device": "cpu"},
|
39 |
+
encode_kwargs={"normalize_embeddings": True},
|
40 |
+
)
|
41 |
+
|
42 |
+
# --- Unzip Vectorstore if needed ---
|
43 |
+
zip_path = "faiss_index.zip"
|
44 |
+
extract_path = "faiss_index"
|
45 |
+
if not os.path.exists(extract_path):
|
46 |
+
with zipfile.ZipFile(zip_path, 'r') as z:
|
47 |
+
z.extractall(extract_path)
|
48 |
+
print("✅ Unzipped FAISS index.")
|
49 |
+
|
50 |
+
# --- Load FAISS Vectorstore & create retriever ---
|
51 |
+
vectorstore = FAISS.load_local(
|
52 |
+
extract_path,
|
53 |
+
embeddings,
|
54 |
+
allow_dangerous_deserialization=True
|
55 |
+
)
|
56 |
+
retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
|
57 |
+
print("✅ FAISS index loaded.")
|
58 |
+
|
59 |
+
# --- Prepare prompt template ---
|
60 |
+
quiz_solving_prompt = """
|
61 |
+
You are an Arabic Hadith Finder assistant.
|
62 |
+
Your goal is to provide an accurate and concise answer extracted directly from the provided retrieved context.
|
63 |
+
Your task is to output only the exact Arabic Hadith (as it appears in the context), removing any extraneous or irrelevant data.
|
64 |
+
|
65 |
+
Instructions:
|
66 |
+
1. Identify the segment in the retrieved context that directly answers the user's question.
|
67 |
+
2. Output the Hadith exactly as it appears in Arabic in the context.
|
68 |
+
3. Remove any information that does not pertain directly to the query.
|
69 |
+
4. If the context does not contain sufficient information to answer the question, respond with "لا أعلم". Do not add or infer any extra information.
|
70 |
+
5. Provide the complete reference of the Hadith (if available), including:
|
71 |
+
- Chapter Number and Name (Arabic and/or English)
|
72 |
+
- Section Number and Name
|
73 |
+
- Hadith Number
|
74 |
+
- Arabic Isnad and Matn
|
75 |
+
- Arabic Grade (if present)
|
76 |
+
- Hadith Book name
|
77 |
+
|
78 |
+
Retrieved context:
|
79 |
+
{context}
|
80 |
+
|
81 |
+
User's question:
|
82 |
+
{question}
|
83 |
+
|
84 |
+
Your response:
|
85 |
+
"""
|
86 |
+
prompt = PromptTemplate(
|
87 |
+
template=quiz_solving_prompt,
|
88 |
+
input_variables=["context", "question"]
|
89 |
+
)
|
90 |
+
|
91 |
+
# --- Assemble a stateless RetrievalQA chain (no memory) ---
|
92 |
+
chain = RetrievalQA.from_chain_type(
|
93 |
+
llm=llm,
|
94 |
+
chain_type="stuff",
|
95 |
+
retriever=retriever,
|
96 |
+
return_source_documents=False,
|
97 |
+
chain_type_kwargs={"prompt": prompt},
|
98 |
+
verbose=False,
|
99 |
+
)
|
100 |
+
|
101 |
+
@app.get("/")
|
102 |
+
def root():
|
103 |
+
return {"message": "Arabic Hadith Finder API is up..."}
|
104 |
+
|
105 |
+
@app.post("/query")
|
106 |
+
def query(request: QueryRequest):
|
107 |
+
try:
|
108 |
+
result = chain.invoke({"query": request.question})
|
109 |
+
return {"answer": result["result"]}
|
110 |
+
except Exception as e:
|
111 |
+
raise HTTPException(status_code=500, detail=str(e))
|