pages/1_🔥_WarmMolGen.py

import streamlit as st
import streamlit.components.v1 as components
import pandas as pd
import mols2grid
import textwrap
import numpy as np
from transformers import EncoderDecoderModel, RobertaTokenizer

# @st.cache(allow_output_mutation=False, hash_funcs={Tokenizer: str})
@st.cache(suppress_st_warning=True)
def load_models():
    # protein_tokenizer = RobertaTokenizer.from_pretrained("gokceuludogan/WarmMolGenTwo")
    # mol_tokenizer = RobertaTokenizer.from_pretrained("seyonec/PubChem10M_SMILES_BPE_450k")
    model1 = EncoderDecoderModel.from_pretrained("gokceuludogan/WarmMolGenOne")
    model2 = EncoderDecoderModel.from_pretrained("gokceuludogan/WarmMolGenTwo")
    return model1, model2#, protein_tokenizer, mol_tokenizer


def warmmolgen_demo():
    protein_tokenizer = RobertaTokenizer.from_pretrained("gokceuludogan/WarmMolGenTwo")
    mol_tokenizer = RobertaTokenizer.from_pretrained("seyonec/PubChem10M_SMILES_BPE_450k")
    #model1, model2, protein_tokenizer, mol_tokenizer = load_models()
    model1, model2 = load_models()

    st.sidebar.subheader("Configurable parameters")

    model_name = st.sidebar.selectbox(
        "Model Selector",
        options=[
            "WarmMolGenOne",
            "WarmMolGenTwo",
        ],
        index=0,
    )

    num_mols = st.sidebar.number_input(
        "Number of generated molecules",
        min_value=0,
        max_value=20,
        value=10,
        help="The number of molecules to be generated.",
    )

    max_new_tokens = st.sidebar.number_input(
        "Maximum length",
        min_value=0,
        max_value=1024,
        value=128,
        help="The maximum length of the sequence to be generated.",
    )
    # temp = st.sidebar.slider(
    #     "Temperature",
    #     value=1.0,
    #     min_value=0.1,
    #     max_value=100.0,
    #     help="The value used to module the next token probabilities.",
    # )
    # top_k = st.sidebar.number_input(
    #     "Top k",
    #     value=10,
    #     help="The number of highest probability vocabulary tokens to keep for top-k-filtering.",
    # )
    # top_p = st.sidebar.number_input(
    #     "Top p",
    #     value=0.95,
    #     help=" If set to float < 1, only the most probable tokens with probabilities that add up to top_p or higher are kept for generation.",
    # )
    do_sample = st.sidebar.selectbox(
        "Sampling?",
        (True, False),
        help="Whether or not to use sampling; use beam decoding otherwise.",
    )
    # num_beams = st.sidebar.number_input(
    #     "Number of beams",
    #     min_value=0,
    #     max_value=20,
    #     value=0,
    #     help="The number of beams to use for beam search.",
    # )
    num_beams = None if do_sample is True else int(num_mols)
    # repetition_penalty = st.sidebar.number_input(
    #     "Repetition Penalty",
    #     min_value=0.0,
    #     value=3.0,
    #     step=0.1,
    #     help="The parameter for repetition penalty. 1.0 means no penalty",
    # )
    # no_repeat_ngram_size = st.sidebar.number_input(
    #     "No Repeat N-Gram Size",
    #     min_value=0,
    #     value=3,
    #     help="If set to int > 0, all ngrams of that size can only occur once.",
    # )
    target = st.text_area(
        "Target Sequence",
        "MENTENSVDSKSIKNLEPKIIHGSESMDSGISLDNSYKMDYPEMGLCIIINNKNFHKSTG",
    )
    inputs = protein_tokenizer(target, return_tensors="pt")

    model = model1 if model_name == 'WarmMolGenOne' else model2
    outputs = model.generate(**inputs, decoder_start_token_id=mol_tokenizer.bos_token_id, 
                             eos_token_id=mol_tokenizer.eos_token_id, pad_token_id=mol_tokenizer.eos_token_id, 
                             max_length=int(max_new_tokens), num_return_sequences=int(num_mols), 
                             do_sample=do_sample, num_beams=num_beams)
    output_smiles = mol_tokenizer.batch_decode(outputs, skip_special_tokens=True)
    st.write("### Generated Molecules")
    #st.write(output_smiles)    
    df_smiles = pd.DataFrame({'SMILES': output_smiles})
    #st.write(df_smiles)    
    raw_html = mols2grid.display(df_smiles, mapping={"SMILES": "SMILES"})._repr_html_()
    components.html(raw_html, width=900, height=450, scrolling=True)
    st.markdown("## How to Generate")
    generation_code = f"""
    from transformers import EncoderDecoderModel, RobertaTokenizer

    protein_tokenizer = RobertaTokenizer.from_pretrained("gokceuludogan/{model_name}") 
    mol_tokenizer = RobertaTokenizer.from_pretrained("seyonec/PubChem10M_SMILES_BPE_450k")
    model = EncoderDecoderModel.from_pretrained("gokceuludogan/{model_name}") 

    inputs = protein_tokenizer("{target}", return_tensors="pt")
    outputs = model.generate(**inputs, decoder_start_token_id=mol_tokenizer.bos_token_id, 
                             eos_token_id=mol_tokenizer.eos_token_id, pad_token_id=mol_tokenizer.eos_token_id, 
                             max_length={max_new_tokens}, num_return_sequences={num_mols}, do_sample={do_sample}, num_beams={num_beams})

    mol_tokenizer.batch_decode(outputs, skip_special_tokens=True)
    """
    st.code(textwrap.dedent(generation_code)) # textwrap.dedent("".join("Halletcez")))

st.set_page_config(page_title="WarmMolGen Demo", page_icon="🔥", layout='wide')
st.markdown("# WarmMolGen Demo")
st.sidebar.header("WarmMolGen Demo")
st.markdown(
    """
    This demo illustrates WarmMolGen models' generation capabilities. 


    Given a target sequence and a set of parameters, the models generate molecules targeting the given protein sequence.


    Please enter an input sequence below 👇  and configure parameters from the sidebar 👈 to generate molecules!


    See below for saving the output molecules and the code snippet generating them!
"""
)

warmmolgen_demo()