Streamlit Value for Multiselect not updating

Wassup Guys,
I am working on an app for keyword prediction (table).
I have a multi select dropdown, where you can select the loaded words you want to replace, if they are inaccurate. I have a helper method (top_k), that returns me a frame of data containing name and ids.
However, when I want to load the new keys from the new table, when a user replaced a key, it works.
I tried to append the new keys into an existing array, that I use as the value for the dropdown, with success.

But the multi-select is not updating the new values I received and added to the word list (array).
Nevertheless I can see my new my session state keys including the old ones, when I print the values of the array.

What did I miss? (Below is a short screen cast where you can see, that the new loaded keys are not included in my multi select)

#checkbox if triggered
retrain = column_2.checkbox('🔂 Retrain model')
            #arr for values
            wordList = []
            #arr for sel val in multi sel
            selectedWords = []
            #save loaded keys in list
            for key in top_k['name']:

#if triggered retrain without sel keys
            if retrain:
                retrainKeys = column_2.checkbox('🔀 Replace chosen keys')

                if "options" not in st.session_state:
                    st.session_state['keys'] = wordList
#set session state for multi select using exisiting arr

                if 'sel_key' in st.session_state:
                        f'⏏️ Chosen Words: {st.session_state["sel_key"]}')
#print key

                    "", st.session_state['keys'], key='sel_key')
#appen session vals and keys for watcher

#if retrained
                if retrainKeys:
                    with st.spinner('Reload model...'):
                        for key in st.session_state["sel_key"]:
                            print('SELECTED: ', selectedWords)
#new top k names
                        retrained_top_k = retrainModel(selectedWords,
                                                       input_text, label_embeddings, st.session_state.model, st.session_state.tokenizer, n_keywords)
                        for key in retrained_top_k['name']:
#add into wordList and sort
                            print('KEEEEEY: ', key)
                            wordList = list(dict.fromkeys(wordList))
                        st.session_state["keys"] = wordList

                        column_1.dataframe(retrained_top_k, height=500)

Example Video