app.py

from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from transformers import pipeline
import torch

main = FastAPI()

# CORS සක්‍රීය කිරීම - ඕනෑම තැනක සිට සම්බන්ධ වීමට
main.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

MODEL_ID = "tencent/Hy-MT1.5-1.8B-2bit"
print(f"🔱 Specialist, Loading {MODEL_ID} (Public Mode)...")

# Pipeline initialization
pipe = pipeline(
    "text-generation",
    model=MODEL_ID,
    device_map="cpu",
    model_kwargs={"trust_remote_code": True}
)

class ChatRequest(BaseModel):
    message: str

@main.post("/v1/chat")
async def chat(request_data: ChatRequest):
    user_query = request_data.message.strip()
    
    # Prompt Setup
    prompt = f"User: {user_query}\nAssistant:"

    # Text Generation
    results = pipe(
        prompt, 
        max_new_tokens=150, 
        do_sample=True, 
        temperature=0.7,
        pad_token_id=50256
    )

    # Result Cleaning
    generated_text = results[0]['generated_text']
    reply = generated_text.split("Assistant:")[-1].strip()

    if not reply:
        reply = "I am here, MINZO-PRIME. Systems are nominal."

    return {"reply": reply}

@main.get("/")
def health():
    return {"status": "Public Inachi-Lite Online"}