app.py

import gradio as gr
import os
os.environ["KERAS_BACKEND"] = "tensorflow"                                                                           
import keras
import keras_nlp
import utils

css = """
html, body {
    margin: 0;
    padding: 0;
    height: 100%;
    overflow: hidden;
}

body::before {
    content: '';
    position: fixed;
    top: 0;
    left: 0;
    width: 100vw;
    height: 100vh;
    background-image: url('https://stsci-opo.org/STScI-01J5E849R5W27ZZ2C3QAE9ET75.png');
    background-size: cover;
    background-repeat: no-repeat;
    opacity: 0.35;             /* Faint background image */
    background-position: center;
    z-index: -1;    /* Keep the background behind text */
}
.gradio-container {
    display: flex;
    justify-content: center;
    align-items: center;
    height: 100vh;  /* Ensure the content is vertically centered */
}
"""

#keras.config.set_floatx("bfloat16") uncomment to run at half precision.
gemma_lm = keras_nlp.models.CausalLM.from_preset("hf://sultan-hassan/CosmoGemma_2b_en")
#chat = utils.ChatState(gemma_lm) # adding a chat helper to manage the conversation state (modified version from https://github.com/google-gemini/gemma-cookbook/blob/main/Gemma/Keras_Gemma_2_Quickstart_Chat.ipynb)

examples = [
    'What are some common methods for model selection in astrophysics?', 
    'What could be the cause of remaining excess power in a signal after foreground mitigation?',
    'What type of coupling in inflationary models can affect the prediction of inflationary parameters?',
    'Please list two ideas for interesting projects to work on to advance cosmology?',
    'Please answer the following multiple-choice question. 
    Question: What is the primary goal of calibrating subgrid feedback models in cosmological simulations?
    A) To ensure that simulations accurately reproduce the observed properties of the interstellar medium.
    B) To create a diverse range of galaxy morphologies in the simulations.
    C) To achieve convergence in simulation results across different resolutions and box sizes.
    D) To steer simulations towards producing a broadly realistic galaxy population that is consistent with key observational constraints.
    Give only one answer, either A, B, C, or D, but not more than one, and always give an answer.',
]

def launch(message):

    # Uncomment for QA system without chat history/memory
    #
    template = "Instruction:\n{instruction}\n\nResponse:\n{response}"
    prompt = template.format(
        instruction=message,                                                                   
        response="",
    )
    out = gemma_lm.generate(prompt, max_length=1024)
    ind = out.index('\n\nResponse:\n') + len('\n\nResponse:\n')
    return out[ind:]
    
    # This is to convert QA system to a chatbot by adding history to prompt
    #chat_history = chat.send_message(message) + "\n\n##### History #####\n\n" +  chat.get_history()  
    #return chat_history

iface = gr.Interface(launch,
                     inputs="text",
                     outputs="text",
                     examples=examples,
                     css=css,
                     title="Hey I am CosmoGemma 👋 I can answer cosmology questions from astroph.CO research articles. Try me :)",
                     description="Gemma_2b_en fine-tuned on QA pairs (~3.5k) generated from Cosmology and Nongalactic Astrophysics articles (arXiv astro-ph.CO) from 2018-2022 and tested on QA pairs (~1k) generated from 2023 articles, scoring over 75% accuracy.")

iface.launch()