main.py

from llama_cpp import Llama
from fastapi import FastAPI
from pydantic import BaseModel
import requests
from ctransformers import AutoModelForCausalLM

llm = AutoModelForCausalLM.from_pretrained("TinyLlama/TinyLlama-1.1B-Chat-v0.6", model_file="ggml-model-q4_0.gguf")

#Pydantic object
class validation(BaseModel):
    prompt: str
#Fast API
app = FastAPI()

@app.post("/llm_on_cpu")
async def stream(item: validation):
    prefix="""<|user|>
    """
    suffix="""<|endoftext|><|assistant|>"""
    user="""
    {prompt}"""

    prompt = f"{prefix}{user.replace('{prompt}', item.prompt)}{suffix}"
    return llm(prompt)