Merge pull request #58 from barun-saha/byok

app.py

@@ -5,7 +5,6 @@ import datetime
 import logging
 import pathlib
 import random
-import sys
 import tempfile
 from typing import List, Union
 
@@ -17,9 +16,6 @@ from langchain_community.chat_message_histories import StreamlitChatMessageHisto
 from langchain_core.messages import HumanMessage
 from langchain_core.prompts import ChatPromptTemplate
 
-sys.path.append('..')
-sys.path.append('../..')
-
 from global_config import GlobalConfig
 from helpers import llm_helper, pptx_helper, text_helper
 
@@ -54,17 +50,58 @@ def _get_prompt_template(is_refinement: bool) -> str:
     return template
 
 
-@st.cache_resource
-def _get_llm(repo_id: str, max_new_tokens: int):
+def are_all_inputs_valid(
+        user_prompt: str,
+        selected_provider: str,
+        selected_model: str,
+        user_key: str,
+) -> bool:
+    """
+    Validate user input and LLM selection.
+
+    :param user_prompt: The prompt.
+    :param selected_provider: The LLM provider.
+    :param selected_model: Name of the model.
+    :param user_key: User-provided API key.
+    :return: `True` if all inputs "look" OK; `False` otherwise.
+    """
+
+    if not text_helper.is_valid_prompt(user_prompt):
+        handle_error(
+            'Not enough information provided!'
+            ' Please be a little more descriptive and type a few words'
+            ' with a few characters :)',
+            False
+        )
+        return False
+
+    if not selected_provider or not selected_model:
+        handle_error('No valid LLM provider and/or model name found!', False)
+        return False
+
+    if not llm_helper.is_valid_llm_provider_model(selected_provider, selected_model, user_key):
+        handle_error(
+            'The LLM settings do not look correct. Make sure that an API key/access token'
+            ' is provided if the selected LLM requires it.',
+            False
+        )
+        return False
+
+    return True
+
+
+def handle_error(error_msg: str, should_log: bool):
     """
-    Get an LLM instance.
+    Display an error message in the app.
 
-    :param repo_id: The model name.
-    :param max_new_tokens: The max new tokens to generate.
-    :return: The LLM.
+    :param error_msg: The error message to be displayed.
+    :param should_log: If `True`, log the message.
     """
 
-    return llm_helper.get_hf_endpoint(repo_id, max_new_tokens)
+    if should_log:
+        logger.error(error_msg)
+
+    st.error(error_msg)
 
 
 APP_TEXT = _load_strings()
@@ -81,18 +118,32 @@ texts = list(GlobalConfig.PPTX_TEMPLATE_FILES.keys())
 captions = [GlobalConfig.PPTX_TEMPLATE_FILES[x]['caption'] for x in texts]
 
 with st.sidebar:
+    # The PPT templates
     pptx_template = st.sidebar.radio(
-        'Select a presentation template:',
+        '1: Select a presentation template:',
         texts,
         captions=captions,
         horizontal=True
     )
-    st.divider()
-    llm_to_use = st.sidebar.selectbox(
-        'Select an LLM to use:',
-        [f'{k} ({v["description"]})' for k, v in GlobalConfig.HF_MODELS.items()]
+
+    # The LLMs
+    llm_provider_to_use = st.sidebar.selectbox(
+        label='2: Select an LLM to use:',
+        options=[f'{k} ({v["description"]})' for k, v in GlobalConfig.VALID_MODELS.items()],
+        index=GlobalConfig.DEFAULT_MODEL_INDEX,
+        help=GlobalConfig.LLM_PROVIDER_HELP,
     ).split(' ')[0]
 
+    # The API key/access token
+    api_key_token = st.text_input(
+        label=(
+            '3: Paste your API key/access token:\n\n'
+            '*Mandatory* for Cohere and Gemini LLMs.'
+            ' *Optional* for HF Mistral LLMs but still encouraged.\n\n'
+        ),
+        type='password',
+    )
+
 
 def build_ui():
     """
@@ -119,119 +170,108 @@ def set_up_chat_ui():
     with st.expander('Usage Instructions'):
         st.markdown(GlobalConfig.CHAT_USAGE_INSTRUCTIONS)
 
-    st.info(
-        'If you like SlideDeck AI, please consider leaving a heart ❤️ on the'
-        ' [Hugging Face Space](https://huggingface.co/spaces/barunsaha/slide-deck-ai/) or'
-        ' a star ⭐ on [GitHub](https://github.com/barun-saha/slide-deck-ai).'
-        ' Your [feedback](https://forms.gle/JECFBGhjvSj7moBx9) is appreciated.'
-    )
-
-    # view_messages = st.expander('View the messages in the session state')
-
-    st.chat_message('ai').write(
-        random.choice(APP_TEXT['ai_greetings'])
-    )
+    st.info(APP_TEXT['like_feedback'])
+    st.chat_message('ai').write(random.choice(APP_TEXT['ai_greetings']))
 
     history = StreamlitChatMessageHistory(key=CHAT_MESSAGES)
-
-    if _is_it_refinement():
-        template = _get_prompt_template(is_refinement=True)
-    else:
-        template = _get_prompt_template(is_refinement=False)
-
-    prompt_template = ChatPromptTemplate.from_template(template)
+    prompt_template = ChatPromptTemplate.from_template(
+        _get_prompt_template(
+            is_refinement=_is_it_refinement()
+        )
+    )
 
     # 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')
+        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
     ):
-        if not text_helper.is_valid_prompt(prompt):
-            st.error(
-                'Not enough information provided!'
-                ' Please be a little more descriptive and type a few words'
-                ' with a few characters :)'
-            )
+        provider, llm_name = llm_helper.get_provider_model(llm_provider_to_use)
+
+        if not are_all_inputs_valid(prompt, provider, llm_name, api_key_token):
             return
 
         logger.info(
             'User input: %s | #characters: %d | LLM: %s',
-            prompt, len(prompt), llm_to_use
+            prompt, len(prompt), llm_name
         )
         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():
+            user_messages = _get_user_messages()
+            user_messages.append(prompt)
+            list_of_msgs = [
+                f'{idx + 1}. {msg}' for idx, msg in enumerate(user_messages)
+            ]
             formatted_template = prompt_template.format(
                 **{
-                    'instructions': list_of_msgs,
+                    'instructions': '\n'.join(list_of_msgs),
                     'previous_content': _get_last_response(),
                 }
             )
         else:
-            formatted_template = prompt_template.format(
-                **{
-                    'question': prompt,
-                }
-            )
+            formatted_template = prompt_template.format(**{'question': prompt})
 
         progress_bar = st.progress(0, 'Preparing to call LLM...')
         response = ''
 
         try:
-            for chunk in _get_llm(
-                    repo_id=llm_to_use,
-                    max_new_tokens=GlobalConfig.HF_MODELS[llm_to_use]['max_new_tokens']
-            ).stream(formatted_template):
-                response += chunk
-
-                # Update the progress bar
-                progress_percentage = min(
-                    len(response) / GlobalConfig.HF_MODELS[llm_to_use]['max_new_tokens'], 0.95
+            llm = llm_helper.get_langchain_llm(
+                provider=provider,
+                model=llm_name,
+                max_new_tokens=GlobalConfig.VALID_MODELS[llm_provider_to_use]['max_new_tokens'],
+                api_key=api_key_token.strip(),
+            )
+
+            if not llm:
+                handle_error(
+                    'Failed to create an LLM instance! Make sure that you have selected the'
+                    ' correct model from the dropdown list and have provided correct API key'
+                    ' or access token.',
+                    False
                 )
+                return
+
+            for _ in llm.stream(formatted_template):
+                response += _
+
+                # Update the progress bar with an approx progress percentage
                 progress_bar.progress(
-                    progress_percentage,
+                    min(
+                        len(response) / GlobalConfig.VALID_MODELS[
+                            llm_provider_to_use
+                        ]['max_new_tokens'],
+                        0.95
+                    ),
                     text='Streaming content...this might take a while...'
                 )
         except requests.exceptions.ConnectionError:
-            msg = (
+            handle_error(
                 'A connection error occurred while streaming content from the LLM endpoint.'
                 ' Unfortunately, the slide deck cannot be generated. Please try again later.'
-                ' Alternatively, try selecting a different LLM from the dropdown list.'
+                ' Alternatively, try selecting a different LLM from the dropdown list.',
+                True
             )
-            logger.error(msg)
-            st.error(msg)
             return
         except huggingface_hub.errors.ValidationError as ve:
-            msg = (
+            handle_error(
                 f'An error occurred while trying to generate the content: {ve}'
-                '\nPlease try again with a significantly shorter input text.'
+                '\nPlease try again with a significantly shorter input text.',
+                True
             )
-            logger.error(msg)
-            st.error(msg)
             return
         except Exception as ex:
-            msg = (
+            handle_error(
                 f'An unexpected error occurred while generating the content: {ex}'
                 '\nPlease try again later, possibly with different inputs.'
                 ' Alternatively, try selecting a different LLM from the dropdown list.'
+                ' If you are using Cohere or Gemini models, make sure that you have provided'
+                ' a correct API key.',
+                True
             )
-            logger.error(msg)
-            st.error(msg)
             return
 
         history.add_user_message(prompt)
@@ -240,25 +280,20 @@ def set_up_chat_ui():
         # 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
-        response_cleaned = text_helper.get_clean_json(response)
-
+        response = text_helper.get_clean_json(response)
         logger.info(
-            'Cleaned JSON response:: original length: %d | cleaned length: %d',
-            len(response), len(response_cleaned)
+            'Cleaned JSON length: %d', len(response)
         )
-        # logger.debug('Cleaned JSON: %s', response_cleaned)
 
         # Now create the PPT file
         progress_bar.progress(
             GlobalConfig.LLM_PROGRESS_MAX,
             text='Finding photos online and generating the slide deck...'
         )
-        path = generate_slide_deck(response_cleaned)
         progress_bar.progress(1.0, text='Done!')
-
         st.chat_message('ai').code(response, language='json')
 
-        if path:
+        if path := generate_slide_deck(response):
             _display_download_button(path)
 
         logger.info(
@@ -279,44 +314,35 @@ def generate_slide_deck(json_str: str) -> Union[pathlib.Path, None]:
     try:
         parsed_data = json5.loads(json_str)
     except ValueError:
-        st.error(
-            'Encountered error while parsing JSON...will fix it and retry'
-        )
-        logger.error(
-            'Caught ValueError: trying again after repairing JSON...'
+        handle_error(
+            'Encountered error while parsing JSON...will fix it and retry',
+            True
         )
         try:
             parsed_data = json5.loads(text_helper.fix_malformed_json(json_str))
         except ValueError:
-            st.error(
+            handle_error(
                 'Encountered an error again while fixing JSON...'
                 'the slide deck cannot be created, unfortunately ☹'
-                '\nPlease try again later.'
-            )
-            logger.error(
-                'Caught ValueError: failed to repair JSON!'
+                '\nPlease try again later.',
+                True
             )
-
             return None
     except RecursionError:
-        st.error(
-            'Encountered an error while parsing JSON...'
+        handle_error(
+            'Encountered a recursion error while parsing JSON...'
             'the slide deck cannot be created, unfortunately ☹'
-            '\nPlease try again later.'
+            '\nPlease try again later.',
+            True
         )
-        logger.error('Caught RecursionError while parsing JSON. Cannot generate the slide deck!')
-
         return None
     except Exception:
-        st.error(
+        handle_error(
             'Encountered an error while parsing JSON...'
             'the slide deck cannot be created, unfortunately ☹'
-            '\nPlease try again later.'
+            '\nPlease try again later.',
+            True
         )
-        logger.error(
-            'Caught ValueError: failed to parse JSON!'
-        )
-
         return None
 
     if DOWNLOAD_FILE_KEY in st.session_state:

global_config.py

@@ -17,16 +17,39 @@ class GlobalConfig:
     A data class holding the configurations.
     """
 
-    HF_MODELS = {
-        'mistralai/Mistral-7B-Instruct-v0.2': {
+    PROVIDER_COHERE = 'co'
+    PROVIDER_GOOGLE_GEMINI = 'gg'
+    PROVIDER_HUGGING_FACE = 'hf'
+    VALID_PROVIDERS = {PROVIDER_COHERE, PROVIDER_GOOGLE_GEMINI, PROVIDER_HUGGING_FACE}
+    VALID_MODELS = {
+        '[co]command-r-08-2024': {
+            'description': 'simpler, slower',
+            'max_new_tokens': 4096,
+            'paid': True,
+        },
+        '[gg]gemini-1.5-flash-002': {
+            'description': 'faster response',
+            'max_new_tokens': 8192,
+            'paid': True,
+        },
+        '[hf]mistralai/Mistral-7B-Instruct-v0.2': {
             'description': 'faster, shorter',
-            'max_new_tokens': 8192
+            'max_new_tokens': 8192,
+            'paid': False,
         },
-        'mistralai/Mistral-Nemo-Instruct-2407': {
+        '[hf]mistralai/Mistral-Nemo-Instruct-2407': {
             'description': 'longer response',
-            'max_new_tokens': 12228
+            'max_new_tokens': 10240,
+            'paid': False,
         },
     }
+    LLM_PROVIDER_HELP = (
+        'LLM provider codes:\n\n'
+        '- **[co]**: Cohere\n'
+        '- **[gg]**: Google Gemini API\n'
+        '- **[hf]**: Hugging Face Inference Endpoint\n'
+    )
+    DEFAULT_MODEL_INDEX = 2
     LLM_MODEL_TEMPERATURE = 0.2
     LLM_MODEL_MIN_OUTPUT_LENGTH = 100
     LLM_MODEL_MAX_INPUT_LENGTH = 400  # characters

helpers/llm_helper.py

@@ -1,18 +1,28 @@
+"""
+Helper functions to access LLMs.
+"""
 import logging
+import re
+import sys
+from typing import Tuple, Union
+
 import requests
 from requests.adapters import HTTPAdapter
 from urllib3.util import Retry
+from langchain_core.language_models import BaseLLM
 
-from langchain_community.llms.huggingface_endpoint import HuggingFaceEndpoint
-from langchain_core.language_models import LLM
+sys.path.append('..')
 
 from global_config import GlobalConfig
 
 
-HF_API_HEADERS = {"Authorization": f"Bearer {GlobalConfig.HUGGINGFACEHUB_API_TOKEN}"}
+LLM_PROVIDER_MODEL_REGEX = re.compile(r'\[(.*?)\](.*)')
+HF_API_HEADERS = {'Authorization': f'Bearer {GlobalConfig.HUGGINGFACEHUB_API_TOKEN}'}
 REQUEST_TIMEOUT = 35
 
 logger = logging.getLogger(__name__)
+logging.getLogger('httpx').setLevel(logging.WARNING)
+logging.getLogger('httpcore').setLevel(logging.WARNING)
 
 retries = Retry(
     total=5,
@@ -27,101 +37,123 @@ http_session.mount('https://', adapter)
 http_session.mount('http://', adapter)
 
 
-def get_hf_endpoint(repo_id: str, max_new_tokens: int) -> LLM:
+def get_provider_model(provider_model: str) -> Tuple[str, str]:
     """
-    Get an LLM via the HuggingFaceEndpoint of LangChain.
+    Parse and get LLM provider and model name from strings like `[provider]model/name-version`.
 
-    :param repo_id: The model name.
-    :param max_new_tokens: The max new tokens to generate.
-    :return: The HF LLM inference endpoint.
+    :param provider_model: The provider, model name string from `GlobalConfig`.
+    :return: The provider and the model name.
     """
 
-    logger.debug('Getting LLM via HF endpoint: %s', repo_id)
-
-    return HuggingFaceEndpoint(
-        repo_id=repo_id,
-        max_new_tokens=max_new_tokens,
-        top_k=40,
-        top_p=0.95,
-        temperature=GlobalConfig.LLM_MODEL_TEMPERATURE,
-        repetition_penalty=1.03,
-        streaming=True,
-        huggingfacehub_api_token=GlobalConfig.HUGGINGFACEHUB_API_TOKEN,
-        return_full_text=False,
-        stop_sequences=['</s>'],
-    )
-
-
-# def hf_api_query(payload: dict) -> dict:
-#     """
-#     Invoke HF inference end-point API.
-#
-#     :param payload: The prompt for the LLM and related parameters.
-#     :return: The output from the LLM.
-#     """
-#
-#     try:
-#         response = http_session.post(
-#             HF_API_URL,
-#             headers=HF_API_HEADERS,
-#             json=payload,
-#             timeout=REQUEST_TIMEOUT
-#         )
-#         result = response.json()
-#     except requests.exceptions.Timeout as te:
-#         logger.error('*** Error: hf_api_query timeout! %s', str(te))
-#         result = []
-#
-#     return result
-
-
-# def generate_slides_content(topic: str) -> str:
-#     """
-#     Generate the outline/contents of slides for a presentation on a given topic.
-#
-#     :param topic: Topic on which slides are to be generated.
-#     :return: The content in JSON format.
-#     """
-#
-#     with open(GlobalConfig.SLIDES_TEMPLATE_FILE, 'r', encoding='utf-8') as in_file:
-#         template_txt = in_file.read().strip()
-#         template_txt = template_txt.replace('<REPLACE_PLACEHOLDER>', topic)
-#
-#     output = hf_api_query({
-#         'inputs': template_txt,
-#         '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
-#         }
-#     })
-#
-#     output = output[0]['generated_text'].strip()
-#     # output = output[len(template_txt):]
-#
-#     json_end_idx = output.rfind('```')
-#     if json_end_idx != -1:
-#         # logging.debug(f'{json_end_idx=}')
-#         output = output[:json_end_idx]
-#
-#     logger.debug('generate_slides_content: output: %s', output)
-#
-#     return output
+    match = LLM_PROVIDER_MODEL_REGEX.match(provider_model)
 
+    if match:
+        inside_brackets = match.group(1)
+        outside_brackets = match.group(2)
+        return inside_brackets, outside_brackets
 
-if __name__ == '__main__':
-    # results = get_related_websites('5G AI WiFi 6')
-    #
-    # for a_result in results.results:
-    #     print(a_result.title, a_result.url, a_result.extract)
+    return '', ''
+
+
+def is_valid_llm_provider_model(provider: str, model: str, api_key: str) -> bool:
+    """
+    Verify whether LLM settings are proper.
+    This function does not verify whether `api_key` is correct. It only confirms that the key has
+    at least five characters. Key verification is done when the LLM is created.
+
+    :param provider: Name of the LLM provider.
+    :param model: Name of the model.
+    :param api_key: The API key or access token.
+    :return: `True` if the settings "look" OK; `False` otherwise.
+    """
+
+    if not provider or not model or provider not in GlobalConfig.VALID_PROVIDERS:
+        return False
+
+    if provider in [GlobalConfig.PROVIDER_GOOGLE_GEMINI, ]:
+        if not api_key or len(api_key) < 5:
+            return False
+
+    return True
 
-    # get_ai_image('A talk on AI, covering pros and cons')
-    pass
+
+def get_langchain_llm(
+        provider: str,
+        model: str,
+        max_new_tokens: int,
+        api_key: str = ''
+) -> Union[BaseLLM, None]:
+    """
+    Get an LLM based on the provider and model specified.
+
+    :param provider: The LLM provider. Valid values are `hf` for Hugging Face.
+    :param model: The name of the LLM.
+    :param max_new_tokens: The maximum number of tokens to generate.
+    :param api_key: API key or access token to use.
+    :return: An instance of the LLM or `None` in case of any error.
+    """
+
+    if provider == GlobalConfig.PROVIDER_HUGGING_FACE:
+        from langchain_community.llms.huggingface_endpoint import HuggingFaceEndpoint
+
+        logger.debug('Getting LLM via HF endpoint: %s', model)
+        return HuggingFaceEndpoint(
+            repo_id=model,
+            max_new_tokens=max_new_tokens,
+            top_k=40,
+            top_p=0.95,
+            temperature=GlobalConfig.LLM_MODEL_TEMPERATURE,
+            repetition_penalty=1.03,
+            streaming=True,
+            huggingfacehub_api_token=api_key or GlobalConfig.HUGGINGFACEHUB_API_TOKEN,
+            return_full_text=False,
+            stop_sequences=['</s>'],
+        )
+
+    if provider == GlobalConfig.PROVIDER_GOOGLE_GEMINI:
+        from google.generativeai.types.safety_types import HarmBlockThreshold, HarmCategory
+        from langchain_google_genai import GoogleGenerativeAI
+
+        logger.debug('Getting LLM via Google Gemini: %s', model)
+        return GoogleGenerativeAI(
+            model=model,
+            temperature=GlobalConfig.LLM_MODEL_TEMPERATURE,
+            max_tokens=max_new_tokens,
+            timeout=None,
+            max_retries=2,
+            google_api_key=api_key,
+            safety_settings={
+                HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT:
+                    HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE,
+                HarmCategory.HARM_CATEGORY_HATE_SPEECH: HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE,
+                HarmCategory.HARM_CATEGORY_HARASSMENT: HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE,
+                HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT:
+                    HarmBlockThreshold.BLOCK_LOW_AND_ABOVE
+            }
+        )
+
+    if provider == GlobalConfig.PROVIDER_COHERE:
+        from langchain_cohere.llms import Cohere
+
+        logger.debug('Getting LLM via Cohere: %s', model)
+        return Cohere(
+            temperature=GlobalConfig.LLM_MODEL_TEMPERATURE,
+            max_tokens=max_new_tokens,
+            timeout_seconds=None,
+            max_retries=2,
+            cohere_api_key=api_key,
+            streaming=True,
+        )
+
+    return None
+
+
+if __name__ == '__main__':
+    inputs = [
+        '[co]Cohere',
+        '[hf]mistralai/Mistral-7B-Instruct-v0.2',
+        '[gg]gemini-1.5-flash-002'
+    ]
+
+    for text in inputs:
+        print(get_provider_model(text))

helpers/pptx_helper.py

@@ -115,37 +115,42 @@ def generate_powerpoint_presentation(
 
     # Add content in a loop
     for a_slide in parsed_data['slides']:
-        is_processing_done = _handle_icons_ideas(
-            presentation=presentation,
-            slide_json=a_slide,
-            slide_width_inch=slide_width_inch,
-            slide_height_inch=slide_height_inch
-        )
-
-        if not is_processing_done:
-            is_processing_done = _handle_double_col_layout(
+        try:
+            is_processing_done = _handle_icons_ideas(
                 presentation=presentation,
                 slide_json=a_slide,
                 slide_width_inch=slide_width_inch,
                 slide_height_inch=slide_height_inch
             )
 
-        if not is_processing_done:
-            is_processing_done = _handle_step_by_step_process(
-                presentation=presentation,
-                slide_json=a_slide,
-                slide_width_inch=slide_width_inch,
-                slide_height_inch=slide_height_inch
-            )
+            if not is_processing_done:
+                is_processing_done = _handle_double_col_layout(
+                    presentation=presentation,
+                    slide_json=a_slide,
+                    slide_width_inch=slide_width_inch,
+                    slide_height_inch=slide_height_inch
+                )
 
-        if not is_processing_done:
-            _handle_default_display(
-                presentation=presentation,
-                slide_json=a_slide,
-                slide_width_inch=slide_width_inch,
-                slide_height_inch=slide_height_inch
+            if not is_processing_done:
+                is_processing_done = _handle_step_by_step_process(
+                    presentation=presentation,
+                    slide_json=a_slide,
+                    slide_width_inch=slide_width_inch,
+                    slide_height_inch=slide_height_inch
+                )
+
+            if not is_processing_done:
+                _handle_default_display(
+                    presentation=presentation,
+                    slide_json=a_slide,
+                    slide_width_inch=slide_width_inch,
+                    slide_height_inch=slide_height_inch
             )
 
+        except Exception:
+            # In case of any unforeseen error, try to salvage what is available
+            continue
+
     # The thank-you slide
     last_slide_layout = presentation.slide_layouts[0]
     slide = presentation.slides.add_slide(last_slide_layout)

helpers/text_helper.py

@@ -1,3 +1,6 @@
+"""
+Utility functions to help with text processing.
+"""
 import json_repair as jr
 
 
@@ -17,28 +20,19 @@ def is_valid_prompt(prompt: str) -> bool:
 
 def get_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.
+    Attempt to clean a JSON response string from the LLM by removing ```json at the beginning and
+    trailing ``` and any text beyond that.
     CAUTION: May not be always accurate.
 
     :param json_str: The input string in JSON format.
     :return: The "cleaned" JSON string.
     """
 
-    # An example of response containing JSON and other text:
-    # {
-    #     "title": "AI and the Future: A Transformative Journey",
-    #     "slides": [
-    #       ...
-    #     ]
-    # }    <<---- This is end of valid JSON content
-    # ```
-    #
-    # ```vbnet
-    # Please note that the JSON output is in valid format but the content of the "Role of GPUs in AI" slide is just an example and may not be factually accurate. For accurate information, you should consult relevant resources and update the content accordingly.
-    # ```
     response_cleaned = json_str
 
+    if json_str.startswith('```json'):
+        json_str = json_str[7:]
+
     while True:
         idx = json_str.rfind('```')  # -1 on failure
 
@@ -46,7 +40,7 @@ def get_clean_json(json_str: str) -> str:
             break
 
         # In the ideal scenario, the character before the last ``` should be
-        # a new line or a closing bracket }
+        # a new line or a closing bracket
         prev_char = json_str[idx - 1]
 
         if (prev_char == '}') or (prev_char == '\n' and json_str[idx - 2] == '}'):
@@ -69,13 +63,13 @@ def fix_malformed_json(json_str: str) -> str:
 
 
 if __name__ == '__main__':
-    json1 = '''{
+    JSON1 = '''{
     "key": "value"
     }
     '''
-    json2 = '''["Reason": "Regular updates help protect against known vulnerabilities."]'''
-    json3 = '''["Reason" Regular updates help protect against known vulnerabilities."]'''
-    json4 = '''
+    JSON2 = '''["Reason": "Regular updates help protect against known vulnerabilities."]'''
+    JSON3 = '''["Reason" Regular updates help protect against known vulnerabilities."]'''
+    JSON4 = '''
     {"bullet_points": [
         ">> Write without stopping or editing",
         >> Set daily writing goals and stick to them,
@@ -83,7 +77,7 @@ if __name__ == '__main__':
     ],}
     '''
 
-    print(fix_malformed_json(json1))
-    print(fix_malformed_json(json2))
-    print(fix_malformed_json(json3))
-    print(fix_malformed_json(json4))
+    print(fix_malformed_json(JSON1))
+    print(fix_malformed_json(JSON2))
+    print(fix_malformed_json(JSON3))
+    print(fix_malformed_json(JSON4))

requirements.txt

@@ -10,6 +10,8 @@ pydantic==2.9.1
 langchain~=0.3.7
 langchain-core~=0.3.0
 langchain-community==0.3.0
+langchain-google-genai==2.0.6
+langchain-cohere==0.3.3
 streamlit~=1.38.0
 
 python-pptx
@@ -19,7 +21,6 @@ requests~=2.32.3
 
 transformers~=4.44.0
 torch==2.4.0
-langchain-community
 
 urllib3~=2.2.1
 lxml~=4.9.3

strings.json

@@ -33,5 +33,6 @@
         "Looks like you have a looming deadline. Can I help you get started with your slide deck?",
         "Hello! What topic do you have on your mind today?"
     ],
-    "chat_placeholder": "Write the topic or instructions here"
+    "chat_placeholder": "Write the topic or instructions here",
+    "like_feedback": "If you like SlideDeck AI, please consider leaving a heart ❤\uFE0F on the [Hugging Face Space](https://huggingface.co/spaces/barunsaha/slide-deck-ai/) or a star ⭐ on [GitHub](https://github.com/barun-saha/slide-deck-ai). Your [feedback](https://forms.gle/JECFBGhjvSj7moBx9) is appreciated."
 }