# from ibm_watsonx_ai.foundation_models import ModelInference
# from ibm_watsonx_ai.metanames import GenTextParamsMetaNames as GenParams
# from ibm_watsonx_ai.metanames import EmbedTextParamsMetaNames
# from ibm_watsonx_ai import Credentials
# from langchain_ibm import WatsonxLLM, WatsonxEmbeddings
# from langchain.text_splitter import RecursiveCharacterTextSplitter
# from langchain_text_splitters import RecursiveCharacterTextSplitter
import uuid
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Chroma
from langchain_community.document_loaders import PyPDFLoader
# from langchain.chains import RetrievalQA
from langchain_classic.chains import RetrievalQA
from langchain_community.llms import HuggingFacePipeline
from langchain_community.embeddings import HuggingFaceEmbeddings
# from langchain_huggingface import HuggingFacePipeline
import torch

from huggingface_hub import HfFolder
from huggingface_hub import hf_hub_download
from langchain_community.llms import LlamaCpp
import gradio as gr

from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline, BitsAndBytesConfig

## You can use this section to suppress warnings generated by your code:
def warn(*args, **kwargs):
    pass
import warnings
warnings.warn = warn
warnings.filterwarnings('ignore')

## LLM
# def get_llm():
#     model_id = 'ibm/granite-3-2-8b-instruct'
#     parameters = {
#         GenParams.MAX_NEW_TOKENS: 256, # This controls the maximum number of tokens in the generated output
#         GenParams.TEMPERATURE: 0.5, # This controls randomness or creativity of the model's responses
#     }
#     project_id = 'skills-network'
#     watsonx_llm = WatsonxLLM(
#         model_id = model_id,
#         url="https://us-south.ml.cloud.ibm.com",
#         project_id = project_id,
#         params = parameters,
#     )
#     return watsonx_llm

LLM_INSTANCE = None

def get_llm():
    global LLM_INSTANCE
    if LLM_INSTANCE is not None:
        return LLM_INSTANCE

    # Download quantized GGUF model (works on CPU)
    model_path = hf_hub_download(
        repo_id="TheBloke/Mistral-7B-Instruct-v0.3-GGUF",
        filename="mistral-7b-instruct-v0.3.Q4_K_M.gguf",
        local_dir=".",
        local_dir_use_symlinks=False,
    )

    llm = LlamaCpp(
        model_path=model_path,
        n_ctx=4096,
        n_threads=4,
        temperature=0.3,
        top_p=0.9,
        verbose=False,
    )

    LLM_INSTANCE = llm
    return llm


## Document loader
def document_loader(file):
    loader = PyPDFLoader(file)
    loaded_document = loader.load()
    return loaded_document
    
## Text splitter
def text_splitter(data):
    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size = 1000,
        chunk_overlap = 50,
        length_function = len,
    )
    chuncks = text_splitter.split_documents(data)
    return chuncks
## Embedding model


# def watsonx_embedding():
#     embed_params = {
#         EmbedTextParamsMetaNames.TRUNCATE_INPUT_TOKENS: 256,
#         EmbedTextParamsMetaNames.RETURN_OPTIONS: {'input_text': True},
#     }
    
#     watsonx_embedding = WatsonxEmbeddings(
#         model_id = "ibm/slate-125m-english-rtrvr-v2",
#         url="https://us-south.ml.cloud.ibm.com",
#         project_id = "skills-network",
#         params = embed_params,
#     )
#     return watsonx_embedding

def watsonx_embedding():
    model_id = 'sentence-transformers/all-MiniLM-L6-v2'
    embedding_model = HuggingFaceEmbeddings(model_name=model_id)
    return embedding_model

## Vector db
def vector_database(chunks):
    
    chunks = [
        c for c in chunks
        if c.page_content and c.page_content.strip() and len(c.page_content.strip()) > 10
    ]
    embedding_model = watsonx_embedding()
    # We add 'collection_name' with a random ID (uuid)
    # This forces Chroma to create a brand new, empty bucket for every PDF
    vectordb = Chroma.from_documents(
        documents=chunks, 
        embedding=embedding_model,
        collection_name=f"pdf_chat_{uuid.uuid4().hex}" 
    )
    return vectordb
    
## Retriever
def retriever(file):
    splits = document_loader(file)
    chunks = text_splitter(splits)
    vectordb = vector_database(chunks)
    retriever = vectordb.as_retriever()
    
    return retriever
    
## QA Chain
def retriever_qa(file, query):
    llm = get_llm()
    retriever_obj = retriever(file)
    qa = RetrievalQA.from_chain_type(
        llm = llm, 
        chain_type = 'stuff',
        retriever = retriever_obj,
        return_source_documents = False
    )
    response = qa.invoke(query)
    return response['result']
    
## Create Gradio interface
rag_application = gr.Interface(
    fn = retriever_qa,
    # allow_flagging = 'never',
    flagging_mode = 'never',
    inputs = [
        gr.File(label = 'Upload PDF File', file_count = 'single', file_types = ['.pdf'], type ='filepath'), # Drag and drop file upload
        gr.Textbox(label= 'Input Query', lines = 2, placeholder = 'Type your question here...')
    ],
    outputs = gr.Textbox(label = 'Ouput'),
    title= 'PDF QA Chatbot',
    description = 'Upload a PDF document and ask any question. The chatbot will try to answer using the provided document.'
)
# Launch the app
rag_application.launch()