app/app.py

# =======================
# ⚡ GGUF + llama-cpp-python FastAPI App for HF Spaces (CPU Optimized)
# =======================

from fastapi import FastAPI, Request
from pydantic import BaseModel
from llama_cpp import Llama
import os
from app.policy_vector_db import PolicyVectorDB, ensure_db_populated

# Initialize FastAPI app
app = FastAPI()

# -----------------------------
# ✅ Vector DB Configuration
# -----------------------------
DB_PERSIST_DIRECTORY = "/app/vector_database"
CHUNKS_FILE_PATH = "/app/processed_chunks.json"

print("[INFO] Initializing vector DB...")
db = PolicyVectorDB(persist_directory=DB_PERSIST_DIRECTORY)
if not ensure_db_populated(db, CHUNKS_FILE_PATH):
    print("[WARNING] DB not populated. Chunks file may be missing.")
else:
    print("[INFO] Vector DB ready.")

# -----------------------------
# ✅ Load GGUF Model with llama-cpp-python (model is pre-downloaded in Dockerfile)
# -----------------------------
MODEL_PATH = "/app/tinyllama_dop_q4_k_m.gguf"

print(f"[INFO] Loading GGUF model from: {MODEL_PATH}")
llm = Llama(
    model_path=MODEL_PATH,
    n_ctx=1024,
    n_threads=2,
    n_batch=32,
    use_mlock=False,
    verbose=False
)
print("[INFO] Model loaded successfully.")

# -----------------------------
# ✅ Request Schema
# -----------------------------
class Query(BaseModel):
    question: str

# -----------------------------
# ✅ Chat Endpoint
# -----------------------------
@app.post("/chat/")
async def chat(query: Query):
    question = query.question
    search_results = db.search(question)
    context = "\n".join([res["text"] for res in search_results])

    prompt = f"""### Context:\n{context}\n\n### Question: {question}\n### Answer:"""

    response = llm(prompt, max_tokens=150, stop=["###"])
    answer = response["choices"][0]["text"].strip()

    return {"answer": answer}