main.py

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

llms = {
"tinyllama":{"name": "TheBloke/TinyLlama-1.1B-1T-OpenOrca-GGUF", "file":"tinyllama-1.1b-1t-openorca.Q4_K_M.gguf", "suffix":"<|im_end|><|im_start|>assistant", "prefix":"<|im_start|>system You are a helpful assistant  <|im_end|><|im_start|>user"},
"tinyllama2":{"name": "TheBloke/TinyLlama-1.1B-1T-OpenOrca-GGUF", "file":"tinyllama-1.1b-1t-openorca.Q3_K_M.gguf", "suffix":"<|im_end|><|im_start|>assistant", "prefix":"<|im_start|>system You are a helpful assistant  <|im_end|><|im_start|>user"},
"tinyllama3":{"name": "TheBloke/TinyLlama-1.1B-Chat-v0.3-GGUF", "file":"tinyllama-1.1b-chat-v0.3.Q4_K_M.gguf", "suffix":"<|im_end|><|im_start|>assistant", "prefix":"<|im_start|>system You are a helpful assistant  <|im_end|><|im_start|>user"},
"tinyllama4":{"name": "TheBloke/TinyLlama-1.1B-Chat-v0.3-GGUF", "file":"tinyllama-1.1b-chat-v0.3.Q2_K.gguf", "suffix":"<|im_end|><|im_start|>assistant", "prefix":"<|im_start|>system You are a helpful assistant  <|im_end|><|im_start|>user"},
"solar":{"name": "TheBloke/SOLAR-10.7B-Instruct-v1.0-GGUF", "file":"solar-10.7b-instruct-v1.0.Q4_K_M.gguf", "suffix":"\n### Assistant:\n", "prefix":"### User:\n"},
"phi2":{"name": "TheBloke/phi-2-GGUF", "file":"phi-2.Q4_K_M.gguf", "suffix":"\nOutput:\n", "prefix":"Instruct:\n"}
}

for k in llms.keys():
    AutoModelForCausalLM.from_pretrained(llms[k]['name'], model_file=llms[k]['file'])

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

@app.post("/llm_on_cpu")
async def stream(item: validation):

    prefix=llms[item.llm]['prefix']
    suffix=llms[item.llm]['suffix']
    user="""
    {prompt}"""
    
    llm = AutoModelForCausalLM.from_pretrained(llms[item.llm]['name'], model_file=llms[item.llm]['file'])

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