Rename new chat app and legacy app files

app.py

@@ -1,282 +1,395 @@
-import pathlib
+import datetime
 import logging
+import pathlib
+import random
 import tempfile
-from typing import List, Tuple
+from typing import List
 
 import json5
-import metaphor_python as metaphor
 import streamlit as st
+from langchain_community.chat_message_histories import (
+    StreamlitChatMessageHistory
+)
+from langchain_core.messages import HumanMessage
+from langchain_core.prompts import ChatPromptTemplate
+from transformers import AutoTokenizer
 
-from helpers import llm_helper, pptx_helper
 from global_config import GlobalConfig
+from helpers import llm_helper, pptx_helper
 
 
-APP_TEXT = json5.loads(open(GlobalConfig.APP_STRINGS_FILE, 'r', encoding='utf-8').read())
-GB_CONVERTER = 2 ** 30
-
+@st.cache_data
+def _load_strings() -> dict:
+    """
+    Load various strings to be displayed in the app.
+    :return: The dictionary of strings.
+    """
 
-logger = logging.getLogger(__name__)
+    with open(GlobalConfig.APP_STRINGS_FILE, 'r', encoding='utf-8') as in_file:
+        return json5.loads(in_file.read())
 
 
 @st.cache_data
-def get_contents_wrapper(text: str) -> str:
+def _get_prompt_template(is_refinement: bool) -> str:
     """
-    Fetch and cache the slide deck contents on a topic by calling an external API.
+    Return a prompt template.
 
-    :param text: The presentation topic.
-    :return: The slide deck contents or outline in JSON format.
+    :param is_refinement: Whether this is the initial or refinement prompt.
+    :return: The prompt template as f-string.
     """
 
-    logger.info('LLM call because of cache miss...')
-    return llm_helper.generate_slides_content(text).strip()
+    if is_refinement:
+        with open(GlobalConfig.REFINEMENT_PROMPT_TEMPLATE, 'r', encoding='utf-8') as in_file:
+            template = in_file.read()
+    else:
+        with open(GlobalConfig.INITIAL_PROMPT_TEMPLATE, 'r', encoding='utf-8') as in_file:
+            template = in_file.read()
+
+    return template
 
 
 @st.cache_resource
-def get_metaphor_client_wrapper() -> metaphor.Metaphor:
+def _get_tokenizer() -> AutoTokenizer:
     """
-    Create a Metaphor client for semantic Web search.
+    Get Mistral tokenizer for counting tokens.
 
-    :return: Metaphor instance.
+    :return: The tokenizer.
     """
 
-    return metaphor.Metaphor(api_key=GlobalConfig.METAPHOR_API_KEY)
+    return AutoTokenizer.from_pretrained(
+        pretrained_model_name_or_path=GlobalConfig.HF_LLM_MODEL_NAME
+    )
 
 
-@st.cache_data
-def get_web_search_results_wrapper(text: str) -> List[Tuple[str, str]]:
-    """
-    Fetch and cache the Web search results on a given topic.
+APP_TEXT = _load_strings()
 
-    :param text: The topic.
-    :return: A list of (title, link) tuples.
-    """
-
-    results = []
-    search_results = get_metaphor_client_wrapper().search(
-        text,
-        use_autoprompt=True,
-        num_results=5
-    )
+# Session variables
+CHAT_MESSAGES = 'chat_messages'
+DOWNLOAD_FILE_KEY = 'download_file_name'
+IS_IT_REFINEMENT = 'is_it_refinement'
 
-    for a_result in search_results.results:
-        results.append((a_result.title, a_result.url))
+logger = logging.getLogger(__name__)
+progress_bar = st.progress(0, text='Setting up SlideDeck AI...')
 
-    return results
+texts = list(GlobalConfig.PPTX_TEMPLATE_FILES.keys())
+captions = [GlobalConfig.PPTX_TEMPLATE_FILES[x]['caption'] for x in texts]
+pptx_template = st.sidebar.radio(
+    'Select a presentation template:',
+    texts,
+    captions=captions,
+    horizontal=True
+)
 
 
-def build_ui():
+def display_page_header_content():
     """
-    Display the input elements for content generation. Only covers the first step.
+    Display content in the page header.
     """
 
-    # get_disk_used_percentage()
-
     st.title(APP_TEXT['app_name'])
     st.subheader(APP_TEXT['caption'])
-    st.markdown(
-        'Powered by'
-        ' [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2).'
-    )
-    st.markdown(
-        '*If the JSON is generated or parsed incorrectly, try again later by making minor changes'
-        ' to the input text.*'
-    )
+    # st.markdown(
+    #     '![Visitors](https://api.visitorbadge.io/api/visitors?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2Fbarunsaha%2Fslide-deck-ai&countColor=%23263759)'  # noqa: E501
+    # )
 
-    with st.form('my_form'):
-        # Topic input
-        try:
-            with open(GlobalConfig.PRELOAD_DATA_FILE, 'r', encoding='utf-8') as in_file:
-                preload_data = json5.loads(in_file.read())
-        except (FileExistsError, FileNotFoundError):
-            preload_data = {'topic': '', 'audience': ''}
-
-        topic = st.text_area(
-            APP_TEXT['input_labels'][0],
-            value=preload_data['topic']
-        )
 
-        texts = list(GlobalConfig.PPTX_TEMPLATE_FILES.keys())
-        captions = [GlobalConfig.PPTX_TEMPLATE_FILES[x]['caption'] for x in texts]
-
-        pptx_template = st.radio(
-            'Select a presentation template:',
-            texts,
-            captions=captions,
-            horizontal=True
-        )
+def display_page_footer_content():
+    """
+    Display content in the page footer.
+    """
 
-        st.divider()
-        submit = st.form_submit_button('Generate slide deck')
+    st.text(APP_TEXT['tos'] + '\n\n' + APP_TEXT['tos2'])
 
-    if submit:
-        # st.write(f'Clicked {time.time()}')
-        st.session_state.submitted = True
 
-    # https://github.com/streamlit/streamlit/issues/3832#issuecomment-1138994421
-    if 'submitted' in st.session_state:
-        progress_text = 'Generating the slides...give it a moment'
-        progress_bar = st.progress(0, text=progress_text)
+def build_ui():
+    """
+    Display the input elements for content generation.
+    """
 
-        topic_txt = topic.strip()
-        generate_presentation(topic_txt, pptx_template, progress_bar)
+    display_page_header_content()
 
-    st.divider()
-    st.text(APP_TEXT['tos'])
-    st.text(APP_TEXT['tos2'])
+    with st.expander('Usage Policies and Limitations'):
+        display_page_footer_content()
 
-    st.markdown(
-        '![Visitors]'
-        '(https://api.visitorbadge.io/api/visitors?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2Fbarunsaha%2Fslide-deck-ai&countColor=%23263759)'
-    )
+    progress_bar.progress(50, text='Setting up chat interface...')
+    set_up_chat_ui()
 
 
-def generate_presentation(topic: str, pptx_template: str, progress_bar):
+def set_up_chat_ui():
     """
-    Process the inputs to generate the slides.
-
-    :param topic: The presentation topic based on which contents are to be generated.
-    :param pptx_template: The PowerPoint template name to be used.
-    :param progress_bar: Progress bar from the page.
+    Prepare the chat interface and related functionality.
     """
 
-    topic_length = len(topic)
-    logger.debug('Input length:: topic: %s', topic_length)
-
-    if topic_length >= 10:
-        logger.debug('Topic: %s', topic)
-        target_length = min(topic_length, GlobalConfig.LLM_MODEL_MAX_INPUT_LENGTH)
+    with st.expander('Usage Instructions'):
+        st.write(GlobalConfig.CHAT_USAGE_INSTRUCTIONS)
+        st.markdown(
+            'SlideDeck AI is powered by'
+            ' [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2)'
+        )
 
-        try:
-            # Step 1: Generate the contents in JSON format using an LLM
-            json_str = process_slides_contents(topic[:target_length], progress_bar)
-            logger.debug('Truncated topic: %s', topic[:target_length])
-            logger.debug('Length of JSON: %d', len(json_str))
+    # view_messages = st.expander('View the messages in the session state')
 
-            # Step 2: Generate the slide deck based on the template specified
-            if len(json_str) > 0:
-                st.info(
-                    'Tip: The generated content doesn\'t look so great?'
-                    ' Need alternatives? Just change your description text and try again.',
-                    icon="💡️"
-                )
-            else:
-                st.error(
-                    'Unfortunately, JSON generation failed, so the next steps would lead'
-                    ' to nowhere. Try again or come back later.'
-                )
-                return
+    st.chat_message('ai').write(
+        random.choice(APP_TEXT['ai_greetings'])
+    )
+    progress_bar.progress(100, text='Done!')
+    progress_bar.empty()
 
-            all_headers = generate_slide_deck(json_str, pptx_template, progress_bar)
+    history = StreamlitChatMessageHistory(key=CHAT_MESSAGES)
 
-            # Step 3: Bonus stuff: Web references and AI art
-            show_bonus_stuff(all_headers)
+    if _is_it_refinement():
+        template = _get_prompt_template(is_refinement=True)
+        logger.debug('Getting refinement template')
+    else:
+        template = _get_prompt_template(is_refinement=False)
+        logger.debug('Getting initial template')
+
+    prompt_template = ChatPromptTemplate.from_template(template)
+
+    # Since Streamlit app reloads at every interaction, display the chat history
+    # from the save session state
+    for msg in history.messages:
+        msg_type = msg.type
+        if msg_type == 'user':
+            st.chat_message(msg_type).write(msg.content)
+        else:
+            st.chat_message(msg_type).code(msg.content, language='json')
+
+    if prompt := st.chat_input(
+        placeholder=APP_TEXT['chat_placeholder'],
+        max_chars=GlobalConfig.LLM_MODEL_MAX_INPUT_LENGTH
+    ):
+
+        progress_bar_pptx = st.progress(0, 'Preparing to run...')
+        if not _is_valid_prompt(prompt):
+            return
+
+        logger.info('User input: %s', prompt)
+        st.chat_message('user').write(prompt)
+
+        user_messages = _get_user_messages()
+        user_messages.append(prompt)
+        list_of_msgs = [
+            f'{idx + 1}. {msg}' for idx, msg in enumerate(user_messages)
+        ]
+        list_of_msgs = '\n'.join(list_of_msgs)
+
+        if _is_it_refinement():
+            formatted_template = prompt_template.format(
+                **{
+                    'instructions': list_of_msgs,
+                    'previous_content': _get_last_response()
+                }
+            )
+        else:
+            formatted_template = prompt_template.format(
+                **{
+                    'question': prompt,
+                }
+            )
+
+        progress_bar_pptx.progress(5, 'Calling LLM...will retry if connection times out...')
+        response: dict = llm_helper.hf_api_query({
+            'inputs': formatted_template,
+            'parameters': {
+                'temperature': GlobalConfig.LLM_MODEL_TEMPERATURE,
+                'min_length': GlobalConfig.LLM_MODEL_MIN_OUTPUT_LENGTH,
+                'max_length': GlobalConfig.LLM_MODEL_MAX_OUTPUT_LENGTH,
+                'max_new_tokens': GlobalConfig.LLM_MODEL_MAX_OUTPUT_LENGTH,
+                'num_return_sequences': 1,
+                'return_full_text': False,
+                # "repetition_penalty": 0.0001
+            },
+            'options': {
+                'wait_for_model': True,
+
+                'use_cache': True
+            }
+        })
+
+        if len(response) > 0 and 'generated_text' in response[0]:
+            response: str = response[0]['generated_text'].strip()
+
+        st.chat_message('ai').code(response, language='json')
+
+        history.add_user_message(prompt)
+        history.add_ai_message(response)
+
+        if GlobalConfig.COUNT_TOKENS:
+            tokenizer = _get_tokenizer()
+            tokens_count_in = len(tokenizer.tokenize(formatted_template))
+            tokens_count_out = len(tokenizer.tokenize(response))
+            logger.debug(
+                'Tokens count:: input: %d, output: %d',
+                tokens_count_in, tokens_count_out
+            )
+
+        # _display_messages_history(view_messages)
+
+        # The content has been generated as JSON
+        # There maybe trailing ``` at the end of the response -- remove them
+        # To be careful: ``` may be part of the content as well when code is generated
+        progress_bar_pptx.progress(50, 'Analyzing response...')
+        response_cleaned = _clean_json(response)
+
+        # Now create the PPT file
+        progress_bar_pptx.progress(75, 'Creating the slide deck...give it a moment')
+        generate_slide_deck(response_cleaned)
+        progress_bar_pptx.progress(100, text='Done!')
+
+
+def generate_slide_deck(json_str: str):
+    """
+    Create a slide deck.
 
-        except ValueError as ve:
-            st.error(f'Unfortunately, an error occurred: {ve}! '
-                     f'Please change the text, try again later, or report it, sharing your inputs.')
+    :param json_str: The content in *valid* JSON format.
+    """
 
+    if DOWNLOAD_FILE_KEY in st.session_state:
+        path = pathlib.Path(st.session_state[DOWNLOAD_FILE_KEY])
+        logger.debug('DOWNLOAD_FILE_KEY found in session')
     else:
-        st.error('Not enough information provided! Please be little more descriptive :)')
+        temp = tempfile.NamedTemporaryFile(delete=False, suffix='.pptx')
+        path = pathlib.Path(temp.name)
+        st.session_state[DOWNLOAD_FILE_KEY] = str(path)
+        logger.debug('DOWNLOAD_FILE_KEY not found in session')
 
+    logger.debug('Creating PPTX file: %s...', st.session_state[DOWNLOAD_FILE_KEY])
 
-def process_slides_contents(text: str, progress_bar: st.progress) -> str:
-    """
-    Convert given text into structured data and display. Update the UI.
+    try:
+        pptx_helper.generate_powerpoint_presentation(
+            json_str,
+            slides_template=pptx_template,
+            output_file_path=path
+        )
+
+        _display_download_button(path)
+    except ValueError as ve:
+        st.error(APP_TEXT['json_parsing_error'])
+        logger.error('%s', APP_TEXT['json_parsing_error'])
+        logger.error('Additional error info: %s', str(ve))
+    except Exception as ex:
+        st.error(APP_TEXT['content_generation_error'])
+        logger.error('Caught a generic exception: %s', str(ex))
 
-    :param text: The topic description for the presentation.
-    :param progress_bar: Progress bar for this step.
-    :return: The contents as a JSON-formatted string.
+
+def _is_valid_prompt(prompt: str) -> bool:
     """
+    Verify whether user input satisfies the concerned constraints.
 
-    json_str = ''
+    :param prompt: The user input text.
+    :return: True if all criteria are satisfied; False otherwise.
+    """
 
-    try:
-        logger.info('Calling LLM for content generation on the topic: %s', text)
-        json_str = get_contents_wrapper(text)
-    except Exception as ex:
+    if len(prompt) < 5 or ' ' not in prompt:
         st.error(
-            f'An exception occurred while trying to convert to JSON. It could be because of heavy'
-            f' traffic or something else. Try doing it again or try again later.'
-            f'\nError message: {ex}'
+            'Not enough information provided!'
+            ' Please be a little more descriptive and type a few words with a few characters :)'
         )
+        return False
+
+    return True
+
+
+def _is_it_refinement() -> bool:
+    """
+    Whether it is the initial prompt or a refinement.
+
+    :return: True if it is the initial prompt; False otherwise.
+    """
 
-    progress_bar.progress(50, text='Contents generated')
+    if IS_IT_REFINEMENT in st.session_state:
+        return True
 
-    with st.expander('The generated contents (in JSON format)'):
-        st.code(json_str, language='json')
+    if len(st.session_state[CHAT_MESSAGES]) >= 2:
+        # Prepare for the next call
+        st.session_state[IS_IT_REFINEMENT] = True
+        return True
 
-    return json_str
+    return False
 
 
-def generate_slide_deck(json_str: str, pptx_template: str, progress_bar) -> List:
+def _get_user_messages() -> List[str]:
     """
-    Create a slide deck.
+    Get a list of user messages submitted until now from the session state.
 
-    :param json_str: The contents in JSON format.
-    :param pptx_template: The PPTX template name.
-    :param progress_bar: Progress bar.
-    :return: A list of all slide headers and the title.
+    :return: The list of user messages.
     """
 
-    progress_text = 'Creating the slide deck...give it a moment'
-    progress_bar.progress(75, text=progress_text)
+    return [
+        msg.content for msg in st.session_state[CHAT_MESSAGES] if isinstance(msg, HumanMessage)
+    ]
 
-    # # Get a unique name for the file to save -- use the session ID
-    # ctx = st_sr.get_script_run_ctx()
-    # session_id = ctx.session_id
-    # timestamp = time.time()
-    # output_file_name = f'{session_id}_{timestamp}.pptx'
 
-    temp = tempfile.NamedTemporaryFile(delete=False, suffix='.pptx')
-    path = pathlib.Path(temp.name)
+def _get_last_response() -> str:
+    """
+    Get the last response generated by AI.
 
-    logger.info('Creating PPTX file...')
-    all_headers = pptx_helper.generate_powerpoint_presentation(
-        json_str,
-        slides_template=pptx_template,
-        output_file_path=path
-    )
-    progress_bar.progress(100, text='Done!')
+    :return: The response text.
+    """
+
+    return st.session_state[CHAT_MESSAGES][-1].content
 
-    with open(path, 'rb') as f:
-        st.download_button('Download PPTX file', f, file_name='Presentation.pptx')
 
-    return all_headers
+def _display_messages_history(view_messages: st.expander):
+    """
+    Display the history of messages.
+
+    :param view_messages: The list of AI and Human messages.
+    """
 
+    with view_messages:
+        view_messages.json(st.session_state[CHAT_MESSAGES])
 
-def show_bonus_stuff(ppt_headers: List[str]):
+def _clean_json(json_str: str) -> str:
     """
-    Show bonus stuff for the presentation.
+    Attempt to clean a JSON response string from the LLM by removing the trailing ```
+    and any text beyond that. May not be always accurate.
 
-    :param ppt_headers: A list of the slide headings.
+    :param json_str: The input string in JSON format.
+    :return: The "cleaned" JSON string.
     """
 
-    # Use the presentation title and the slide headers to find relevant info online
-    logger.info('Calling Metaphor search...')
-    ppt_text = ' '.join(ppt_headers)
-    search_results = get_web_search_results_wrapper(ppt_text)
-    md_text_items = []
+    str_len = len(json_str)
+    response_cleaned = json_str
 
-    for (title, link) in search_results:
-        md_text_items.append(f'[{title}]({link})')
+    try:
+        idx = json_str.rindex('```')
+        logger.debug(
+            'Fixing JSON response: str_len: %d, idx of ```: %d',
+            str_len, idx
+        )
 
-    with st.expander('Related Web references'):
-        st.markdown('\n\n'.join(md_text_items))
+        if idx + 3 == str_len:
+            # The response ends with ``` -- most likely the end of JSON response string
+            response_cleaned = json_str[:idx]
+        elif idx + 3 < str_len:
+            # Looks like there are some more content beyond the last ```
+            # In the best case, it would be some additional plain-text response from the LLM
+            # and is unlikely to contain } or ] that are present in JSON
+            if '}' not in json_str[idx + 3:]:  # the remainder of the text
+                response_cleaned = json_str[:idx]
+    except ValueError:
+        # No ``` found
+        pass
 
-    logger.info('Done!')
+    return response_cleaned
 
-    # # Avoid image generation. It costs time and an API call, so just limit to the text generation.
-    # with st.expander('AI-generated image on the presentation topic'):
-    #     logger.info('Calling SDXL for image generation...')
-    #     # img_empty.write('')
-    #     # img_text.write(APP_TEXT['image_info'])
-    #     image = get_ai_image_wrapper(ppt_text)
-    #
-    #     if len(image) > 0:
-    #         image = base64.b64decode(image)
-    #         st.image(image, caption=ppt_text)
-    #         st.info('Tip: Right-click on the image to save it.', icon="💡️")
-    #         logger.info('Image added')
+
+def _display_download_button(file_path: pathlib.Path):
+    """
+    Display a download button to download a slide deck.
+
+    :param file_path: The path of the .pptx file.
+    """
+
+    with open(file_path, 'rb') as download_file:
+        st.download_button(
+            'Download PPTX file ⬇️',
+            data=download_file,
+            file_name='Presentation.pptx',
+            key=datetime.datetime.now()
+        )
 
 
 def main():

chat_app.py

@@ -1,404 +0,0 @@
-import datetime
-import logging
-import pathlib
-import random
-import tempfile
-from typing import List
-
-import json5
-import streamlit as st
-from langchain_community.chat_message_histories import (
-    StreamlitChatMessageHistory
-)
-from langchain_core.messages import HumanMessage
-from langchain_core.prompts import ChatPromptTemplate
-from transformers import AutoTokenizer
-
-from global_config import GlobalConfig
-from helpers import llm_helper, pptx_helper
-
-
-@st.cache_data
-def _load_strings() -> dict:
-    """
-    Load various strings to be displayed in the app.
-    :return: The dictionary of strings.
-    """
-
-    with open(GlobalConfig.APP_STRINGS_FILE, 'r', encoding='utf-8') as in_file:
-        return json5.loads(in_file.read())
-
-
-@st.cache_data
-def _get_prompt_template(is_refinement: bool) -> str:
-    """
-    Return a prompt template.
-
-    :param is_refinement: Whether this is the initial or refinement prompt.
-    :return: The prompt template as f-string.
-    """
-
-    if is_refinement:
-        with open(GlobalConfig.REFINEMENT_PROMPT_TEMPLATE, 'r', encoding='utf-8') as in_file:
-            template = in_file.read()
-    else:
-        with open(GlobalConfig.INITIAL_PROMPT_TEMPLATE, 'r', encoding='utf-8') as in_file:
-            template = in_file.read()
-
-    return template
-
-
-@st.cache_resource
-def _get_tokenizer() -> AutoTokenizer:
-    """
-    Get Mistral tokenizer for counting tokens.
-
-    :return: The tokenizer.
-    """
-
-    return AutoTokenizer.from_pretrained(
-        pretrained_model_name_or_path=GlobalConfig.HF_LLM_MODEL_NAME
-    )
-
-
-APP_TEXT = _load_strings()
-
-# Session variables
-CHAT_MESSAGES = 'chat_messages'
-DOWNLOAD_FILE_KEY = 'download_file_name'
-IS_IT_REFINEMENT = 'is_it_refinement'
-
-logger = logging.getLogger(__name__)
-progress_bar = st.progress(0, text='Setting up SlideDeck AI...')
-
-texts = list(GlobalConfig.PPTX_TEMPLATE_FILES.keys())
-captions = [GlobalConfig.PPTX_TEMPLATE_FILES[x]['caption'] for x in texts]
-pptx_template = st.sidebar.radio(
-    'Select a presentation template:',
-    texts,
-    captions=captions,
-    horizontal=True
-)
-
-
-def display_page_header_content():
-    """
-    Display content in the page header.
-    """
-
-    st.title(APP_TEXT['app_name'])
-    st.subheader(APP_TEXT['caption'])
-    # st.markdown(
-    #     '![Visitors](https://api.visitorbadge.io/api/visitors?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2Fbarunsaha%2Fslide-deck-ai&countColor=%23263759)'  # noqa: E501
-    # )
-
-
-def display_page_footer_content():
-    """
-    Display content in the page footer.
-    """
-
-    st.text(APP_TEXT['tos'] + '\n\n' + APP_TEXT['tos2'])
-
-
-def build_ui():
-    """
-    Display the input elements for content generation.
-    """
-
-    display_page_header_content()
-
-    with st.expander('Usage Policies and Limitations'):
-        display_page_footer_content()
-
-    progress_bar.progress(50, text='Setting up chat interface...')
-    set_up_chat_ui()
-
-
-def set_up_chat_ui():
-    """
-    Prepare the chat interface and related functionality.
-    """
-
-    with st.expander('Usage Instructions'):
-        st.write(GlobalConfig.CHAT_USAGE_INSTRUCTIONS)
-        st.markdown(
-            'SlideDeck AI is powered by'
-            ' [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2)'
-        )
-
-    # view_messages = st.expander('View the messages in the session state')
-
-    st.chat_message('ai').write(
-        random.choice(APP_TEXT['ai_greetings'])
-    )
-    progress_bar.progress(100, text='Done!')
-    progress_bar.empty()
-
-    history = StreamlitChatMessageHistory(key=CHAT_MESSAGES)
-
-    if _is_it_refinement():
-        template = _get_prompt_template(is_refinement=True)
-        logger.debug('Getting refinement template')
-    else:
-        template = _get_prompt_template(is_refinement=False)
-        logger.debug('Getting initial template')
-
-    prompt_template = ChatPromptTemplate.from_template(template)
-
-    # Since Streamlit app reloads at every interaction, display the chat history
-    # from the save session state
-    for msg in history.messages:
-        msg_type = msg.type
-        if msg_type == 'user':
-            st.chat_message(msg_type).write(msg.content)
-        else:
-            st.chat_message(msg_type).code(msg.content, language='json')
-
-    if prompt := st.chat_input(
-        placeholder=APP_TEXT['chat_placeholder'],
-        max_chars=GlobalConfig.LLM_MODEL_MAX_INPUT_LENGTH
-    ):
-
-        progress_bar_pptx = st.progress(0, 'Preparing to run...')
-        if not _is_valid_prompt(prompt):
-            return
-
-        logger.info('User input: %s', prompt)
-        st.chat_message('user').write(prompt)
-
-        user_messages = _get_user_messages()
-        user_messages.append(prompt)
-        list_of_msgs = [
-            f'{idx + 1}. {msg}' for idx, msg in enumerate(user_messages)
-        ]
-        list_of_msgs = '\n'.join(list_of_msgs)
-
-        if _is_it_refinement():
-            formatted_template = prompt_template.format(
-                **{
-                    'instructions': list_of_msgs,
-                    'previous_content': _get_last_response()
-                }
-            )
-        else:
-            formatted_template = prompt_template.format(
-                **{
-                    'question': prompt,
-                }
-            )
-
-        progress_bar_pptx.progress(5, 'Calling LLM...will retry if connection times out...')
-        response: dict = llm_helper.hf_api_query({
-            'inputs': formatted_template,
-            'parameters': {
-                'temperature': GlobalConfig.LLM_MODEL_TEMPERATURE,
-                'min_length': GlobalConfig.LLM_MODEL_MIN_OUTPUT_LENGTH,
-                'max_length': GlobalConfig.LLM_MODEL_MAX_OUTPUT_LENGTH,
-                'max_new_tokens': GlobalConfig.LLM_MODEL_MAX_OUTPUT_LENGTH,
-                'num_return_sequences': 1,
-                'return_full_text': False,
-                # "repetition_penalty": 0.0001
-            },
-            'options': {
-                'wait_for_model': True,
-
-                'use_cache': True
-            }
-        })
-
-        if len(response) > 0 and 'generated_text' in response[0]:
-            response: str = response[0]['generated_text'].strip()
-
-        st.chat_message('ai').code(response, language='json')
-
-        history.add_user_message(prompt)
-        history.add_ai_message(response)
-
-        if GlobalConfig.COUNT_TOKENS:
-            tokenizer = _get_tokenizer()
-            tokens_count_in = len(tokenizer.tokenize(formatted_template))
-            tokens_count_out = len(tokenizer.tokenize(response))
-            logger.debug(
-                'Tokens count:: input: %d, output: %d',
-                tokens_count_in, tokens_count_out
-            )
-
-        # _display_messages_history(view_messages)
-
-        # The content has been generated as JSON
-        # There maybe trailing ``` at the end of the response -- remove them
-        # To be careful: ``` may be part of the content as well when code is generated
-        progress_bar_pptx.progress(50, 'Analyzing response...')
-        response_cleaned = _clean_json(response)
-
-        # Now create the PPT file
-        progress_bar_pptx.progress(75, 'Creating the slide deck...give it a moment')
-        generate_slide_deck(response_cleaned)
-        progress_bar_pptx.progress(100, text='Done!')
-
-
-def generate_slide_deck(json_str: str):
-    """
-    Create a slide deck.
-
-    :param json_str: The content in *valid* JSON format.
-    """
-
-    if DOWNLOAD_FILE_KEY in st.session_state:
-        path = pathlib.Path(st.session_state[DOWNLOAD_FILE_KEY])
-        logger.debug('DOWNLOAD_FILE_KEY found in session')
-    else:
-        temp = tempfile.NamedTemporaryFile(delete=False, suffix='.pptx')
-        path = pathlib.Path(temp.name)
-        st.session_state[DOWNLOAD_FILE_KEY] = str(path)
-        logger.debug('DOWNLOAD_FILE_KEY not found in session')
-
-    logger.debug('Creating PPTX file: %s...', st.session_state[DOWNLOAD_FILE_KEY])
-
-    try:
-        pptx_helper.generate_powerpoint_presentation(
-            json_str,
-            slides_template=pptx_template,
-            output_file_path=path
-        )
-
-        _display_download_button(path)
-    except ValueError as ve:
-        st.error(APP_TEXT['json_parsing_error'])
-        logger.error('%s', APP_TEXT['json_parsing_error'])
-        logger.error('Additional error info: %s', str(ve))
-    except Exception as ex:
-        st.error(APP_TEXT['content_generation_error'])
-        logger.error('Caught a generic exception: %s', str(ex))
-
-
-def _is_valid_prompt(prompt: str) -> bool:
-    """
-    Verify whether user input satisfies the concerned constraints.
-
-    :param prompt: The user input text.
-    :return: True if all criteria are satisfied; False otherwise.
-    """
-
-    if len(prompt) < 5 or ' ' not in prompt:
-        st.error(
-            'Not enough information provided!'
-            ' Please be a little more descriptive and type a few words with a few characters :)'
-        )
-        return False
-
-    return True
-
-
-def _is_it_refinement() -> bool:
-    """
-    Whether it is the initial prompt or a refinement.
-
-    :return: True if it is the initial prompt; False otherwise.
-    """
-
-    if IS_IT_REFINEMENT in st.session_state:
-        return True
-
-    if len(st.session_state[CHAT_MESSAGES]) >= 2:
-        # Prepare for the next call
-        st.session_state[IS_IT_REFINEMENT] = True
-        return True
-
-    return False
-
-
-def _get_user_messages() -> List[str]:
-    """
-    Get a list of user messages submitted until now from the session state.
-
-    :return: The list of user messages.
-    """
-
-    return [
-        msg.content for msg in st.session_state[CHAT_MESSAGES] if isinstance(msg, HumanMessage)
-    ]
-
-
-def _get_last_response() -> str:
-    """
-    Get the last response generated by AI.
-
-    :return: The response text.
-    """
-
-    return st.session_state[CHAT_MESSAGES][-1].content
-
-
-def _display_messages_history(view_messages: st.expander):
-    """
-    Display the history of messages.
-
-    :param view_messages: The list of AI and Human messages.
-    """
-
-    with view_messages:
-        view_messages.json(st.session_state[CHAT_MESSAGES])
-
-def _clean_json(json_str: str) -> str:
-    """
-    Attempt to clean a JSON response string from the LLM by removing the trailing ```
-    and any text beyond that. May not be always accurate.
-
-    :param json_str: The input string in JSON format.
-    :return: The "cleaned" JSON string.
-    """
-
-    str_len = len(json_str)
-    response_cleaned = json_str
-
-    try:
-        idx = json_str.rindex('```')
-        logger.debug(
-            'Fixing JSON response: str_len: %d, idx of ```: %d',
-            str_len, idx
-        )
-
-        if idx + 3 == str_len:
-            # The response ends with ``` -- most likely the end of JSON response string
-            response_cleaned = json_str[:idx]
-        elif idx + 3 < str_len:
-            # Looks like there are some more content beyond the last ```
-            # In the best case, it would be some additional plain-text response from the LLM
-            # and is unlikely to contain } or ] that are present in JSON
-            if '}' not in json_str[idx + 3:]:  # the remainder of the text
-                response_cleaned = json_str[:idx]
-    except ValueError:
-        # No ``` found
-        pass
-
-    return response_cleaned
-
-
-def _display_download_button(file_path: pathlib.Path):
-    """
-    Display a download button to download a slide deck.
-
-    :param file_path: The path of the .pptx file.
-    """
-
-    with open(file_path, 'rb') as download_file:
-        st.download_button(
-            'Download PPTX file ⬇️',
-            data=download_file,
-            file_name='Presentation.pptx',
-            key=datetime.datetime.now()
-        )
-
-
-def main():
-    """
-    Trigger application run.
-    """
-
-    build_ui()
-
-
-if __name__ == '__main__':
-    main()

legacy_app.py

@@ -0,0 +1,291 @@
+import pathlib
+import logging
+import tempfile
+from typing import List, Tuple
+
+import json5
+import metaphor_python as metaphor
+import streamlit as st
+
+from helpers import llm_helper, pptx_helper
+from global_config import GlobalConfig
+
+
+APP_TEXT = json5.loads(open(GlobalConfig.APP_STRINGS_FILE, 'r', encoding='utf-8').read())
+GB_CONVERTER = 2 ** 30
+
+
+logger = logging.getLogger(__name__)
+
+
+@st.cache_data
+def get_contents_wrapper(text: str) -> str:
+    """
+    Fetch and cache the slide deck contents on a topic by calling an external API.
+
+    :param text: The presentation topic.
+    :return: The slide deck contents or outline in JSON format.
+    """
+
+    logger.info('LLM call because of cache miss...')
+    return llm_helper.generate_slides_content(text).strip()
+
+
+@st.cache_resource
+def get_metaphor_client_wrapper() -> metaphor.Metaphor:
+    """
+    Create a Metaphor client for semantic Web search.
+
+    :return: Metaphor instance.
+    """
+
+    return metaphor.Metaphor(api_key=GlobalConfig.METAPHOR_API_KEY)
+
+
+@st.cache_data
+def get_web_search_results_wrapper(text: str) -> List[Tuple[str, str]]:
+    """
+    Fetch and cache the Web search results on a given topic.
+
+    :param text: The topic.
+    :return: A list of (title, link) tuples.
+    """
+
+    results = []
+    search_results = get_metaphor_client_wrapper().search(
+        text,
+        use_autoprompt=True,
+        num_results=5
+    )
+
+    for a_result in search_results.results:
+        results.append((a_result.title, a_result.url))
+
+    return results
+
+
+def build_ui():
+    """
+    Display the input elements for content generation. Only covers the first step.
+    """
+
+    # get_disk_used_percentage()
+
+    st.title(APP_TEXT['app_name'])
+    st.subheader(APP_TEXT['caption'])
+    st.markdown(
+        'Powered by'
+        ' [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2).'
+    )
+    st.markdown(
+        '*If the JSON is generated or parsed incorrectly, try again later by making minor changes'
+        ' to the input text.*'
+    )
+
+    with st.form('my_form'):
+        # Topic input
+        try:
+            with open(GlobalConfig.PRELOAD_DATA_FILE, 'r', encoding='utf-8') as in_file:
+                preload_data = json5.loads(in_file.read())
+        except (FileExistsError, FileNotFoundError):
+            preload_data = {'topic': '', 'audience': ''}
+
+        topic = st.text_area(
+            APP_TEXT['input_labels'][0],
+            value=preload_data['topic']
+        )
+
+        texts = list(GlobalConfig.PPTX_TEMPLATE_FILES.keys())
+        captions = [GlobalConfig.PPTX_TEMPLATE_FILES[x]['caption'] for x in texts]
+
+        pptx_template = st.radio(
+            'Select a presentation template:',
+            texts,
+            captions=captions,
+            horizontal=True
+        )
+
+        st.divider()
+        submit = st.form_submit_button('Generate slide deck')
+
+    if submit:
+        # st.write(f'Clicked {time.time()}')
+        st.session_state.submitted = True
+
+    # https://github.com/streamlit/streamlit/issues/3832#issuecomment-1138994421
+    if 'submitted' in st.session_state:
+        progress_text = 'Generating the slides...give it a moment'
+        progress_bar = st.progress(0, text=progress_text)
+
+        topic_txt = topic.strip()
+        generate_presentation(topic_txt, pptx_template, progress_bar)
+
+    st.divider()
+    st.text(APP_TEXT['tos'])
+    st.text(APP_TEXT['tos2'])
+
+    st.markdown(
+        '![Visitors]'
+        '(https://api.visitorbadge.io/api/visitors?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2Fbarunsaha%2Fslide-deck-ai&countColor=%23263759)'
+    )
+
+
+def generate_presentation(topic: str, pptx_template: str, progress_bar):
+    """
+    Process the inputs to generate the slides.
+
+    :param topic: The presentation topic based on which contents are to be generated.
+    :param pptx_template: The PowerPoint template name to be used.
+    :param progress_bar: Progress bar from the page.
+    """
+
+    topic_length = len(topic)
+    logger.debug('Input length:: topic: %s', topic_length)
+
+    if topic_length >= 10:
+        logger.debug('Topic: %s', topic)
+        target_length = min(topic_length, GlobalConfig.LLM_MODEL_MAX_INPUT_LENGTH)
+
+        try:
+            # Step 1: Generate the contents in JSON format using an LLM
+            json_str = process_slides_contents(topic[:target_length], progress_bar)
+            logger.debug('Truncated topic: %s', topic[:target_length])
+            logger.debug('Length of JSON: %d', len(json_str))
+
+            # Step 2: Generate the slide deck based on the template specified
+            if len(json_str) > 0:
+                st.info(
+                    'Tip: The generated content doesn\'t look so great?'
+                    ' Need alternatives? Just change your description text and try again.',
+                    icon="💡️"
+                )
+            else:
+                st.error(
+                    'Unfortunately, JSON generation failed, so the next steps would lead'
+                    ' to nowhere. Try again or come back later.'
+                )
+                return
+
+            all_headers = generate_slide_deck(json_str, pptx_template, progress_bar)
+
+            # Step 3: Bonus stuff: Web references and AI art
+            show_bonus_stuff(all_headers)
+
+        except ValueError as ve:
+            st.error(f'Unfortunately, an error occurred: {ve}! '
+                     f'Please change the text, try again later, or report it, sharing your inputs.')
+
+    else:
+        st.error('Not enough information provided! Please be little more descriptive :)')
+
+
+def process_slides_contents(text: str, progress_bar: st.progress) -> str:
+    """
+    Convert given text into structured data and display. Update the UI.
+
+    :param text: The topic description for the presentation.
+    :param progress_bar: Progress bar for this step.
+    :return: The contents as a JSON-formatted string.
+    """
+
+    json_str = ''
+
+    try:
+        logger.info('Calling LLM for content generation on the topic: %s', text)
+        json_str = get_contents_wrapper(text)
+    except Exception as ex:
+        st.error(
+            f'An exception occurred while trying to convert to JSON. It could be because of heavy'
+            f' traffic or something else. Try doing it again or try again later.'
+            f'\nError message: {ex}'
+        )
+
+    progress_bar.progress(50, text='Contents generated')
+
+    with st.expander('The generated contents (in JSON format)'):
+        st.code(json_str, language='json')
+
+    return json_str
+
+
+def generate_slide_deck(json_str: str, pptx_template: str, progress_bar) -> List:
+    """
+    Create a slide deck.
+
+    :param json_str: The contents in JSON format.
+    :param pptx_template: The PPTX template name.
+    :param progress_bar: Progress bar.
+    :return: A list of all slide headers and the title.
+    """
+
+    progress_text = 'Creating the slide deck...give it a moment'
+    progress_bar.progress(75, text=progress_text)
+
+    # # Get a unique name for the file to save -- use the session ID
+    # ctx = st_sr.get_script_run_ctx()
+    # session_id = ctx.session_id
+    # timestamp = time.time()
+    # output_file_name = f'{session_id}_{timestamp}.pptx'
+
+    temp = tempfile.NamedTemporaryFile(delete=False, suffix='.pptx')
+    path = pathlib.Path(temp.name)
+
+    logger.info('Creating PPTX file...')
+    all_headers = pptx_helper.generate_powerpoint_presentation(
+        json_str,
+        slides_template=pptx_template,
+        output_file_path=path
+    )
+    progress_bar.progress(100, text='Done!')
+
+    with open(path, 'rb') as f:
+        st.download_button('Download PPTX file', f, file_name='Presentation.pptx')
+
+    return all_headers
+
+
+def show_bonus_stuff(ppt_headers: List[str]):
+    """
+    Show bonus stuff for the presentation.
+
+    :param ppt_headers: A list of the slide headings.
+    """
+
+    # Use the presentation title and the slide headers to find relevant info online
+    logger.info('Calling Metaphor search...')
+    ppt_text = ' '.join(ppt_headers)
+    search_results = get_web_search_results_wrapper(ppt_text)
+    md_text_items = []
+
+    for (title, link) in search_results:
+        md_text_items.append(f'[{title}]({link})')
+
+    with st.expander('Related Web references'):
+        st.markdown('\n\n'.join(md_text_items))
+
+    logger.info('Done!')
+
+    # # Avoid image generation. It costs time and an API call, so just limit to the text generation.
+    # with st.expander('AI-generated image on the presentation topic'):
+    #     logger.info('Calling SDXL for image generation...')
+    #     # img_empty.write('')
+    #     # img_text.write(APP_TEXT['image_info'])
+    #     image = get_ai_image_wrapper(ppt_text)
+    #
+    #     if len(image) > 0:
+    #         image = base64.b64decode(image)
+    #         st.image(image, caption=ppt_text)
+    #         st.info('Tip: Right-click on the image to save it.', icon="💡️")
+    #         logger.info('Image added')
+
+
+def main():
+    """
+    Trigger application run.
+    """
+
+    build_ui()
+
+
+if __name__ == '__main__':
+    main()