run.py

#########################################################################################
# Title:  Gradio Interface to LLM-chatbot with RAG-funcionality and ChromaDB on HF-Hub 
# Author: Andreas Fischer
# Date:   December 29th, 2023
# Last update: December 31th, 2023
##########################################################################################


# Chroma-DB
#-----------
import os
import chromadb
dbPath="/home/af/Schreibtisch/gradio/Chroma/db" 
if(os.path.exists(dbPath)==False): 
  dbPath="/home/user/app/db"
print(dbPath)
#client = chromadb.Client()
path=dbPath
client = chromadb.PersistentClient(path=path)
print(client.heartbeat()) 
print(client.get_version())  
print(client.list_collections()) 
from chromadb.utils import embedding_functions
default_ef = embedding_functions.DefaultEmbeddingFunction()
sentence_transformer_ef = embedding_functions.SentenceTransformerEmbeddingFunction(model_name="T-Systems-onsite/cross-en-de-roberta-sentence-transformer")
#instructor_ef = embedding_functions.InstructorEmbeddingFunction(model_name="hkunlp/instructor-large", device="cuda")
print(str(client.list_collections()))

global collection
if("name=ChromaDB1" in str(client.list_collections())):
  print("ChromaDB1 found!")
  collection = client.get_collection(name="ChromaDB1", embedding_function=sentence_transformer_ef)
else:
  print("ChromaDB1 created!")
  collection = client.create_collection(
    "ChromaDB1",
    embedding_function=sentence_transformer_ef,
    metadata={"hnsw:space": "cosine"})
  
  collection.add(
    documents=["The meaning of life is to love.", "This is a sentence", "This is a sentence too"], 
    metadatas=[{"source": "notion"}, {"source": "google-docs"}, {"source": "google-docs"}], 
    ids=["doc1", "doc2", "doc3"], 
  )

print("Database ready!")
print(collection.count()) 


# Model
#-------

from huggingface_hub import InferenceClient
import gradio as gr

client = InferenceClient(
    "mistralai/Mixtral-8x7B-Instruct-v0.1"
    #"mistralai/Mistral-7B-Instruct-v0.1"
)


# Gradio-GUI
#------------

import gradio as gr
import json

def format_prompt(message, history):
  prompt = "<s>"
  #for user_prompt, bot_response in history:
  #  prompt += f"[INST] {user_prompt} [/INST]"
  #  prompt += f" {bot_response}</s> "
  prompt += f"[INST] {message} [/INST]"
  return prompt

def response(
    prompt, history, temperature=0.9, max_new_tokens=500, top_p=0.95, repetition_penalty=1.0,
):
    temperature = float(temperature)
    if temperature < 1e-2: temperature = 1e-2
    top_p = float(top_p)
    generate_kwargs = dict(
        temperature=temperature,
        max_new_tokens=max_new_tokens,
        top_p=top_p,
        repetition_penalty=repetition_penalty,
        do_sample=True,
        seed=42,
    )
    addon=""
    results=collection.query(
      query_texts=[prompt],
      n_results=2,
      #where={"source": "google-docs"}
      #where_document={"$contains":"search_string"}
    )
    dists=["<small>(relevance: "+str(round((1-d)*100)/100)+";" for d in results['distances'][0]]
    sources=["source: "+s["source"]+")</small>" for s in results['metadatas'][0]]
    results=results['documents'][0]
    combination = zip(results,dists,sources)
    combination = [' '.join(triplets) for triplets in combination]
    print(combination)
    if(len(results)>1):
      addon=" Bitte berücksichtige bei deiner Antwort ggf. folgende Auszüge aus unserer Datenbank, sofern sie für die Antwort erforderlich sind. Beantworte die Frage knapp und präzise. Ignoriere unpassende Datenbank-Auszüge OHNE sie zu kommentieren, zu erwähnen oder aufzulisten:\n"+"\n".join(results)
    system="Du bist ein KI-basiertes Assistenzsystem."+addon+"\n\nUser-Anliegen:"   
    #body={"prompt":system+"### Instruktion:\n"+message+"\n\n### Antwort:","max_tokens":500, "echo":"False","stream":"True"} #e.g. SauerkrautLM
    formatted_prompt = format_prompt(system+"\n"+prompt, history)
    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
    output = ""
    for response in stream:
        output += response.token.text
        yield output
    output=output+"\n\n<br><details open><summary><strong>Sources</strong></summary><br><ul>"+ "".join(["<li>" + s + "</li>" for s in combination])+"</ul></details>"
    yield output

gr.ChatInterface(response, chatbot=gr.Chatbot(render_markdown=True),title="German RAG-Interface to the Hugging Face Hub").queue().launch(share=True) #False, server_name="0.0.0.0", server_port=7864)
print("Interface up and running!")