app.py

import json
import subprocess
import requests
from llama_cpp import Llama
import gradio as gr

#url="https://huggingface.co/TheBloke/WizardLM-13B-V1.2-GGUF/resolve/main/wizardlm-13b-v1.2.Q4_0.gguf"
#url="https://huggingface.co/TheBloke/Mixtral-8x7B-Instruct-v0.1-GGUF/resolve/main/mixtral-8x7b-instruct-v0.1.Q4_0.gguf?download=true"
url="https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_0.gguf?download=true"
response = requests.get(url)
with open("./model.gguf", mode="wb") as file:
  file.write(response.content)
print("Model downloaded")

command = ["python3", "-m", "llama_cpp.server", "--model", "./model.gguf", "--host", "0.0.0.0", "--port", "2600"]
subprocess.Popen(command)
print("Model ready!")

#llm = Llama(model_path="./model.gguf")
#def response(input_text, history):
#    output = llm(f"Q: {input_text} A:", max_tokens=256, stop=["Q:", "\n"], echo=True)
#    return output['choices'][0]['text']

def response(message, history):
  #url="https://afischer1985-wizardlm-13b-v1-2-q4-0-gguf.hf.space/v1/completions"
  url="http://0.0.0.0:2600/v1/completions"  
  #body={"prompt":"Im Folgenden findest du eine Instruktion, die eine Aufgabe bescheibt. Schreibe eine Antwort, um die Aufgabe zu lösen.\n\n### Instruktion:\n"+message+"\n\n### Antwort:","max_tokens":500, "echo":"False","stream":"True"}
  #body={"prompt":" chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.\n\nUSER:\n"+message+"\n\nASSISTANT:","max_tokens":500, "echo":"False","stream":"True"}    
  #body={"prompt":system+"### Instruktion:\n"+message+"\n\n### Antwort:","max_tokens":500, "echo":"False","stream":"True"} #e.g. SauerkrautLM
  body={"prompt":"[INST]"+message+"[/INST]","max_tokens":500, "echo":"False","stream":"True"} #e.g. Mixtral-Instruct
  response=""
  buffer=""
  print("URL: "+url)
  print("User: "+message+"\nAI: ")
  for text in requests.post(url, json=body, stream=True):  #-H 'accept: application/json' -H 'Content-Type: application/json'
    if buffer is None: buffer=""
    buffer=str("".join(buffer))
    #print("*** Raw String: "+str(text)+"\n***\n")
    text=text.decode('utf-8')
    if((text.startswith(": ping -")==False) & (len(text.strip("\n\r"))>0)): buffer=buffer+str(text)
    #print("\n*** Buffer: "+str(buffer)+"\n***\n") 
    buffer=buffer.split('"finish_reason": null}]}')
    if(len(buffer)==1):
      buffer="".join(buffer)
      pass
    if(len(buffer)==2):
      part=buffer[0]+'"finish_reason": null}]}'  
      if(part.lstrip('\n\r').startswith("data: ")): part=part.lstrip('\n\r').replace("data: ", "")
      try: 
        part = str(json.loads(part)["choices"][0]["text"])
        print(part, end="", flush=True)
        response=response+part
        buffer="" # reset buffer
      except Exception as e:
        print("Exception:"+str(e))
        pass
    yield response 

gr.ChatInterface(response).queue().launch(share=True) #False, server_name="0.0.0.0", server_port=7864)
print("Interface up and running!")