main.py

from ctransformers import AutoModelForCausalLM
from fastapi import FastAPI, Form
from pydantic import BaseModel

#Model loading
llm = AutoModelForCausalLM.from_pretrained("dolphin-2.0-mistral-7b.Q4_K_S.gguf",
model_type='mistral',
max_new_tokens = 1096,
threads = 3,
)
   

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

#Zephyr completion
@app.post("/llm_on_cpu")
async def stream(item: validation):
    system_prompt = 'Below is an instruction that describes a task. Write a response that appropriately completes the request.'
    start,end = "<|im_start|>", "<|im_end|>"
    prompt = f"<|im_start|>system\n{system_prompt}{end}\n{start}user\n{prompt.strip()}{end}\n"
    return llm(prompt)