app.py

# AI assistant with a RAG system to query information from the CAMELS cosmological simulations using Langchain and deployed with Gradio
# Author: Pablo Villanueva Domingo

from rag import RAG, load_docs
from langchain_community.embeddings import HuggingFaceInstructEmbeddings
from langchain_mistralai import ChatMistralAI
from langchain_core.rate_limiters import InMemoryRateLimiter
import gradio as gr

# Define a limiter to avoid rate limit issues with MistralAI
rate_limiter = InMemoryRateLimiter(
    requests_per_second=0.1,  # <-- MistralAI free. We can only make a request once every second
    check_every_n_seconds=0.01,  # Wake up every 100 ms to check whether allowed to make a request,
    max_bucket_size=10,  # Controls the maximum burst size.
)

# Load the documentation
docs = load_docs()
print("Pages loaded:",len(docs))

# LLM model
llm = ChatMistralAI(model="mistral-large-latest", rate_limiter=rate_limiter)

# Embeddings
embed_model = "sentence-transformers/multi-qa-distilbert-cos-v1"
# embed_model = "nvidia/NV-Embed-v2"
embeddings = HuggingFaceInstructEmbeddings(model_name=embed_model)

# RAG chain
rag_chain = RAG(llm, docs, embeddings)

# Function to handle prompt and query the RAG chain
def handle_prompt(message, history):
    try:
        # Stream output
        out=""
        for chunk in rag_chain.stream(message):
            out += chunk
            yield out
    except:
        raise gr.Error("Requests rate limit exceeded")


if __name__=="__main__":

    # Predefined messages and examples
    description = "AI powered assistant which answers any question related to the [CAMELS simulations](https://www.camel-simulations.org/)."
    greetingsmessage = "Hi, I'm the CAMELS DocBot, I'm here to assist you with any question related to the CAMELS simulations."
    example_questions = [
                        "How can I read a halo file?",
                        "Which simulation suites are included in CAMELS?",
                        "Which are the largest volumes in CAMELS simulations?",
                        "Write a complete snippet of code getting the power spectrum of a simulation"
                        ]

    # Define customized Gradio chatbot
    chatbot = gr.Chatbot([{"role":"assistant", "content":greetingsmessage}],
                        type="messages",
                        avatar_images=["ims/userpic.png","ims/camelslogo.jpg"],
                        height="60vh")

    # Define Gradio interface
    demo = gr.ChatInterface(handle_prompt,
                            type="messages",
                            title="CAMELS DocBot",
                            fill_height=True,
                            examples=example_questions,
                            theme=gr.themes.Soft(),
                            description=description,
                            #cache_examples=False,
                            chatbot=chatbot)

    demo.launch()