app.py

import spaces
import subprocess 

subprocess.run(
    'pip install flash-attn --no-build-isolation',
    env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"},
    shell=True
)


import os
import torch
from dotenv import load_dotenv
from langchain_community.vectorstores import Qdrant
from langchain_huggingface import HuggingFaceEmbeddings
from langchain.prompts import ChatPromptTemplate
from langchain.schema.runnable import RunnablePassthrough
from langchain.schema.output_parser import StrOutputParser
from qdrant_client import QdrantClient, models
from langchain_openai import ChatOpenAI
import gradio as gr
import logging
from typing import List, Tuple
from dataclasses import dataclass
from datetime import datetime
from transformers import AutoTokenizer, AutoModelForCausalLM ,pipeline
from langchain_huggingface.llms import HuggingFacePipeline
import re
from langchain_huggingface.llms import HuggingFacePipeline
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline,BitsAndBytesConfig,TextIteratorStreamer


# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

@dataclass
class Message:
    role: str
    content: str
    timestamp: str

class ChatHistory:
    def __init__(self):
        self.messages: List[Message] = []
    
    def add_message(self, role: str, content: str):
        timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
        self.messages.append(Message(role=role, content=content, timestamp=timestamp))
    
    def get_formatted_history(self, max_messages: int = 10) -> str:
        """Returns the most recent conversation history formatted as a string"""
        recent_messages = self.messages[-max_messages:] if len(self.messages) > max_messages else self.messages
        formatted_history = "\n".join([
            f"{msg.role}: {msg.content}" for msg in recent_messages
        ])
        return formatted_history
    
    def clear(self):
        self.messages = []

# Load environment variables
load_dotenv()

# HuggingFace API Token
HF_TOKEN = os.getenv("HF_TOKEN")
C_apikey = os.getenv("C_apikey")
OPENAPI_KEY = os.getenv("OPENAPI_KEY")

if not HF_TOKEN:
    logger.error("HF_TOKEN is not set in the environment variables.")
    exit(1)

# HuggingFace Embeddings
embeddings = HuggingFaceEmbeddings(model_name="BAAI/bge-large-en-v1.5")

# Qdrant Client Setup
try:
    client = QdrantClient(
        url=os.getenv("QDRANT_URL"),
        api_key=os.getenv("QDRANT_API_KEY"),
        prefer_grpc=False
    )
except Exception as e:
    logger.error("Failed to connect to Qdrant. Ensure QDRANT_URL and QDRANT_API_KEY are correctly set.")
    exit(1)

# Define collection name
collection_name = "mawared"

# Try to create collection
try:
    client.create_collection(
        collection_name=collection_name,
        vectors_config=models.VectorParams(
            size=768,  # GTE-large embedding size
            distance=models.Distance.COSINE
        )
    )
    logger.info(f"Created new collection: {collection_name}")
except Exception as e:
    if "already exists" in str(e):
        logger.info(f"Collection {collection_name} already exists, continuing...")
    else:
        logger.error(f"Error creating collection: {e}")
        exit(1)

# Create Qdrant vector store
db = Qdrant(
    client=client,
    collection_name=collection_name,
    embeddings=embeddings,
)

# Create retriever
retriever = db.as_retriever(
    search_type="similarity",
    search_kwargs={"k": 3}
)


# Load model directly



# Set up the LLM
# llm = ChatOpenAI(
#     base_url="https://api-inference.huggingface.co/v1/",
#     temperature=0,
#     api_key=HF_TOKEN,
#     model="mistralai/Mistral-Nemo-Instruct-2407",
#     max_tokens=None,
#     timeout=None
    
# )


llm = ChatOpenAI(
    # base_url="https://openrouter.ai/api/v1",
    base_url="https://api.cerebras.ai/v1/chat/completions",
    temperature=0,
    api_key=C_apikey,
    model="llama-3.3-70b",
    max_tokens=None,
    timeout=None,
    stream=True
    
)




# quantization_config = BitsAndBytesConfig(
#         load_in_4bit=True,
#         bnb_4bit_compute_dtype=torch.bfloat16,
#         bnb_4bit_quant_type="nf4",
#         bnb_4bit_use_double_quant=True
#     )




# model_id = "unsloth/phi-4"
# tokenizer = AutoTokenizer.from_pretrained(model_id)

# model = AutoModelForCausalLM.from_pretrained(
#         model_id,
#         torch_dtype=torch.float16,
#         device_map="cuda",
#         attn_implementation="flash_attention_2",
#         quantization_config=quantization_config

#     )

# pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, max_new_tokens=8192 )

# llm = HuggingFacePipeline(pipeline=pipe)





# Create prompt template with chat history
template = """
You are an expert assistant specializing in the Mawared HR System. 
Your task is to provide accurate and contextually relevant answers based on the provided context and chat history. 
If you need more information, ask targeted clarifying questions.
Ensure you provide detailed Numbered step by step to the user and be very accurate.
Previous Conversation:
{chat_history}
Current Context:
{context}
Current Question:
{question}
Ask followup questions based on your provided asnwer to create a conversational flow, Only answer form the provided context and chat history , dont make up any information.
answer only and only from the given context and knowledgebase.
Ensure you dont mention where you got the information from and dont mention any pages or documents.
Esnure your answers are detailed and expressive.
Answer:
"""

prompt = ChatPromptTemplate.from_template(template)

# Create the RAG chain with chat history



def create_rag_chain(chat_history: str):
    chain = (
        {
            "context": retriever,
            "question": RunnablePassthrough(),
            "chat_history": lambda x: chat_history
        }
        | prompt
        | llm
        | StrOutputParser()
    )
    return chain

# Initialize chat history
chat_history = ChatHistory()

# Gradio Function
# @spaces.GPU()
def ask_question_gradio(question, history):
    try:
        # Add user question to chat history
        chat_history.add_message("user", question)
        
        # Get formatted history
        formatted_history = chat_history.get_formatted_history()
        
        # Create chain with current chat history
        rag_chain = create_rag_chain(formatted_history)
        
        # Generate response
        response = ""
        for chunk in rag_chain.stream(question):
            response += chunk
        
        # Add assistant response to chat history
        chat_history.add_message("assistant", response)
        
        # Update Gradio chat history
        history.append({"role": "user", "content": question})
        history.append({"role": "assistant", "content": response})
        
        return "", history
    except Exception as e:
        logger.error(f"Error during question processing: {e}")
        return "", history + [{"role": "assistant", "content": "An error occurred. Please try again later."}]

def clear_chat():
    chat_history.clear()
    return [], ""

# Gradio Interface
with gr.Blocks(theme=gr.themes.Soft()) as iface:
    gr.Image("Image.jpg" , width=1200 , height=300 ,show_label=False, show_download_button=False)
    gr.Markdown("# Mawared HR Assistant")
    gr.Markdown('### Instructions')
    gr.Markdown("The first question will always send out an error in chat , try again and the flow should continue normally , its an API issue and we are working on it")

           
    
    chatbot = gr.Chatbot(
        height=400,
        show_label=False,
        type="messages"  # Using the new messages format
    )
    
    with gr.Row():
        question_input = gr.Textbox(
            label="Ask a question:",
            placeholder="Type your question here...",
            scale=25
        )
        clear_button = gr.Button("Clear Chat", scale=1)
    
    question_input.submit(
        ask_question_gradio,
        inputs=[question_input, chatbot],
        outputs=[question_input, chatbot]
    )
    
    clear_button.click(
        clear_chat,
        outputs=[chatbot, question_input]
    )

# Launch the Gradio App
if __name__ == "__main__":
    iface.launch()