import gradio as gr
# from transformers import pipeline
# from transformers.utils import logging
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.vector_stores.chroma import ChromaVectorStore
from llama_index.core import StorageContext
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
import torch
from llama_index.core import VectorStoreIndex
from llama_index.core import Document
from llama_index.core import Settings
from llama_index.llms.huggingface import (
    HuggingFaceInferenceAPI,
    HuggingFaceLLM,
)
from huggingface_hub import login
import chromadb as chromadb
from chromadb.utils import embedding_functions
import shutil
import os
#

last = 0
CHROMA_DATA_PATH = "chroma_data/"
EMBED_MODEL = "BAAI/bge-m3"
# all-MiniLM-L6-v2
CHUNK_SIZE = 800
CHUNK_OVERLAP = 50
max_results = 3
min_len = 40
min_distance = 0.35
max_distance = 0.6
temperature = 0.55
max_tokens=3072
top_p=0.8
frequency_penalty=0.0
presence_penalty=0.15
jezik = "srpski"

system_sr = "Zoveš se U-Chat AI asistent i pomažeš korisniku usluga kompanije United Group. Korisnik postavlja pitanje ili problem, upareno sa dodatnima saznanjima. Na osnovu toga napiši korisniku kratak i ljubazan odgovor koji kompletira njegov zahtev ili mu daje odgovor na pitanje. "
# " Ako ne znaš odgovor, reci da ne znaš, ne izmišljaj ga."
system_sr += "Usluge kompanije United Group uključuju i kablovsku mrežu za digitalnu televiziju, pristup internetu, uređaj EON SMART BOX za TV sadržaj, kao i fiksnu telefoniju."

chroma_client = chromadb.PersistentClient(CHROMA_DATA_PATH)
embedding_func = embedding_functions.SentenceTransformerEmbeddingFunction(
        model_name=EMBED_MODEL
    )

collection = chroma_client.get_or_create_collection(
        name="chroma_data",
        embedding_function=embedding_func,
        metadata={"hnsw:space": "cosine"},
    )
last = collection.count()

#
HF_TOKEN = "wncSKewozDfuZCXCyFbYbAMHgUrfcrumkc"
#
login(token=("hf_" + HF_TOKEN))

system_propmpt = system_sr

# "facebook/blenderbot-400M-distill",  facebook/blenderbot-400M-distill, stabilityai/stablelm-zephyr-3b, BAAI/bge-small-en-v1.5
Settings.llm = HuggingFaceInferenceAPI(model_name="mistralai/Mistral-Nemo-Instruct-2407",
                              device_map="auto",
                              system_prompt = system_propmpt,
                              context_window=4096,
                              max_new_tokens=256,
#                              stopping_ids=[50278, 50279, 50277, 1, 0],
                              generate_kwargs={"temperature": 0.5, "do_sample": False},
#                              tokenizer_kwargs={"max_length": 4096},
                              tokenizer_name="mistralai/Mistral-Nemo-Instruct-2407", 
                             )
# "BAAI/bge-m3"
Settings.embed_model = HuggingFaceEmbedding(model_name="sentence-transformers/all-MiniLM-L6-v2")
#documents = [Document(text="Indian parliament elections happened in April-May 2024. BJP Party won."),
#             Document(text="Indian parliament elections happened in April-May 2021. XYZ Party won."),
#             Document(text="Indian parliament elections happened in 2020. ABC Party won."),
#            ]
#index = VectorStoreIndex.from_documents(
#    documents,
#)
vector_store = ChromaVectorStore(chroma_collection=collection)
index = VectorStoreIndex.from_vector_store(vector_store, embed_model=Settings.embed_model)
query_engine = index.as_query_engine()
def rag(input_text, file):
    if (file):
#        documents = []
#        for f in file:
#            documents += SimpleDirectoryReader(f).load_data()
#        f = file + "*.pdf"
        pathname = os.path.dirname
#        shutil.copyfile(file.name, path)
        print("pathname=", pathname)
        print("basename=", os.path.basename(file))
        print("filename=", file.name)        
        documents = SimpleDirectoryReader(file).load_data()
        index2 = VectorStoreIndex.from_documents(documents)
        query_engine = index2.as_query_engine()
        return query_engine.query(input_text)
#        collection.add(
#            documents=documents,
#        ids=[f"id{last+i}" for i in range(len(documents))],
#        metadatas=[{"state": "s0", "next": "s0", "used": False, "source": 'None', "page": -1, "lang": jezik } for i in range(len(documents)) ]
#    )
    else:
        query_engine = index.as_query_engine()
        return query_engine.query(input_text)

#iface = gr.Interface(fn=rag, inputs=[gr.Textbox(label="Pitanje:", lines=6), gr.File()], 
#                     outputs=[gr.Textbox(label="Odgovor:", lines=6)], 
#                     title="Kako Vam mogu pomoći?",
#                     description= "UChat"
#                    )

def upload_file(filepath):
    name = Path(filepath).name
    documents = SimpleDirectoryReader(file).load_data()
    index = VectorStoreIndex.from_documents(documents)
    query_engine = index.as_query_engine()
    return filepath

with gr.Blocks() as iface:
    gr.Markdown("Uchat")
    file_out = gr.File()
    with gr.Row():
        with gr.Column(scale=1):
            inp = gr.Textbox(label="Pitanje:", lines=6)
            u = gr.UploadButton("Upload a file", file_count="single")
        with gr.Column(scale=1):
            out = gr.Textbox(label="Odgovor:", lines=6)
            sub = gr.Button(label="Pokreni")

    u.upload(upload_file, u, file_out)
    sub.click(rag, inp, out)
    
iface.launch()