app.py

from datetime import datetime
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_parse import LlamaParse
from llama_index.llms.huggingface_api import HuggingFaceInferenceAPI
import os
from dotenv import load_dotenv
import gradio as gr
import markdowm as md
import base64

# Load environment variables
load_dotenv()

# Mapping for display names and actual model names
llm_display_names = {
    "tiiuae/falcon-7b-instruct": "HundAI",
    "mistralai/Mixtral-8x7B-Instruct-v0.1": "Mixtral-8x7B",
    "meta-llama/Meta-Llama-3-8B-Instruct": "Meta-Llama-3",
    "mistralai/Mistral-7B-Instruct-v0.2": "Mistral-7B",
}
embed_models = [
    "BAAI/bge-small-en-v1.5",  # 33.4M
    "NeuML/pubmedbert-base-embeddings",
    "BAAI/llm-embedder",  # 109M
    "BAAI/bge-large-en"  # 335M
]

# Reverse mapping to retrieve original names
llm_reverse_mapping = {v: k for k, v in llm_display_names.items()}

# Update UI to use display names
def set_llm_model(display_name):
    global selected_llm_model_name
    # Retrieve the original model name using the reverse mapping
    selected_llm_model_name = llm_reverse_mapping.get(display_name, display_name)
    print(f"Model selected: {selected_llm_model_name}")

# Respond function remains unchanged
def respond(message, history):
    try:
        # Initialize the LLM with the selected model
        llm = HuggingFaceInferenceAPI(
            model_name=selected_llm_model_name,  # Use the backend model name
            contextWindow=8192,  
            maxTokens=1024,
            temperature=0.3,
            topP=0.9,
            frequencyPenalty=0.5,
            presencePenalty=0.5,
            token=os.getenv("TOKEN")
        )

        # Set up the query engine with the selected LLM
        query_engine = vector_index.as_query_engine(llm=llm)
        bot_message = query_engine.query(message)

        print(f"\n{datetime.now()}:{selected_llm_model_name}:: {message} --> {str(bot_message)}\n")
        return f"{selected_llm_model_name}:\n{str(bot_message)}"
    except Exception as e:
        if str(e) == "'NoneType' object has no attribute 'as_query_engine'":
            return "Please upload a file."
        return f"An error occurred: {e}"

# UI Setup
with gr.Blocks(theme=gr.themes.Soft(font=[gr.themes.GoogleFont("Roboto Mono")]), css='footer {visibility: hidden}') as demo:
    gr.Markdown("")
    with gr.Tabs():
        with gr.TabItem("Introduction"):
            gr.Markdown(md.description)

        with gr.TabItem("Chatbot"):
            with gr.Accordion("IMPORTANT: READ ME FIRST", open=False):
                guid = gr.Markdown(md.guide)
            with gr.Row():
                with gr.Column(scale=1):
                    file_input = gr.File(file_count="single", type='filepath', label="Upload document")
                    embed_model_dropdown = gr.Dropdown(embed_models, label="Select Embedding", interactive=True)
                    with gr.Row():
                        btn = gr.Button("Submit", variant='primary')
                        clear = gr.ClearButton()
                    output = gr.Text(label='Vector Index')
                    
                    # Use display names for LLM dropdown
                    llm_model_dropdown = gr.Dropdown(
                        list(llm_display_names.values()),  # Display names
                        label="Select LLM",
                        interactive=True
                    )
                    
                with gr.Column(scale=3):
                    gr.ChatInterface(
                        fn=respond,
                        chatbot=gr.Chatbot(height=500),
                        show_progress='full',
                        textbox=gr.Textbox(placeholder="Ask me any questions on the uploaded document!", container=False)
                    )
    
    # Set up Gradio interactions
    llm_model_dropdown.change(fn=set_llm_model, inputs=llm_model_dropdown)
    btn.click(fn=load_files, inputs=[file_input, embed_model_dropdown], outputs=output)
    clear.click(lambda: [None] * 3, outputs=[file_input, embed_model_dropdown, output])

# Launch the demo with a public link option
if __name__ == "__main__":
    demo.launch()