Add conversion of slides contents to JSON

app.py

@@ -1,51 +1,83 @@
+import json
+
 import streamlit as st
 
-import llm_agent
+import llm_helper
 from global_config import GlobalConfig
 
 
+UI_BUTTONS = [
+    'Generate slides content',
+    'Generate JSON',
+    'Make the slides'
+]
+
+
 def build_ui():
-    st.write('''
-    # Slides Wizard
-    
-    *Create your next presentation using AI*
-    ''')
-
-    name = st.text_input(
-        f'''**Type in your name**''',
-        value='John Doe'
-    )
+    """
+    Display the input elements for content generation. Only covers the first step.
+    """
+
+    st.title('Slides Wizard')
+    st.subheader('*:blue[Create your next slide deck using AI]*')
+    st.divider()
+
+    st.header('Step 1: Generate your content')
+    st.caption('Let\'s start by generating some contents for your slides')
+
+    # name = st.text_input(
+    #     f'''**Type in your name**''',
+    #     value='John Doe'
+    # )
+
+    try:
+        with open(GlobalConfig.PRELOAD_DATA_FILE, 'r') as in_file:
+            preload_data = json.loads(in_file.read())
+    except (FileExistsError, FileNotFoundError):
+        preload_data = {'topic': '', 'audience': ''}
 
     topic = st.text_area(
         f'''**Describe the topic of the presentation. 
         Avoid mentioning the count of slides.**''',
-        value='''Make a presentation about AI. Talk about its pros, cons, and future prospects. '''
-        '''Add examples of some real-life use cases in engineering and medicine.'''
+        value=preload_data['topic']
     )
 
     audience = st.text_input(
         f'''**Briefly describe your target audience**''',
-        value='I am a teacher and want to present these slides to college students'
+        value=preload_data['audience']
     )
 
-    if st.button('Generate slides'):
+    # Button with callback function
+    st.button(UI_BUTTONS[0], on_click=button_clicked, args=[0])
+
+    if st.session_state.clicked[0]:
         progress_text = 'Generating your presentation slides...give it a moment'
         progress_bar = st.progress(0, text=progress_text)
 
-        name_txt = name.strip()
+        # name_txt = name.strip()
         topic_txt = topic.strip()
         audience_txt = audience.strip()
 
-        process_inputs(name_txt, topic_txt, audience_txt, progress_bar)
+        process_topic_inputs('', topic_txt, audience_txt, progress_bar)
+
 
+def process_topic_inputs(name: str, topic: str, audience: str, progress_bar):
+    """
+    Process the inputs to generate contents for the slides.
 
-def process_inputs(name: str, topic: str, audience: str, progress_bar):
-    name_length = len(name)
+    :param name: Name of the speaker
+    :param topic: The presentation topic
+    :param audience: Target audience description
+    :param progress_bar: Progress bar from the page
+    :return:
+    """
+
+    # name_length = len(name)
     topic_length = len(topic)
     audience_length = len(audience)
-    print(f'Input lengths:: name: {name_length}, topic: {topic_length}, audience: {audience_length}')
+    print(f'Input lengths:: topic: {topic_length}, audience: {audience_length}')
 
-    if name_length > 0 and topic_length > 10 and audience_length > 5:
+    if topic_length > 10 and audience_length > 5:
         print(
             f'Name: {name}\n'
             f'Topic: {topic}\n'
@@ -56,27 +88,72 @@ def process_inputs(name: str, topic: str, audience: str, progress_bar):
         target_length = min(topic_length, GlobalConfig.LLM_MODEL_MAX_INPUT_LENGTH)
 
         try:
-            slides_content = llm_agent.generate_slides_content(name, topic[:target_length], audience)
+            slides_content = llm_helper.generate_slides_content(name, topic[:target_length], audience)
 
             print('=' * 20)
             print(f'Slides content:\n{slides_content}')
             print('=' * 20)
             st.write(f'''Slides content:\n{slides_content}''')
             progress_bar.progress(100, text='Done!')
+
+            # Move on to step 2
+            st.divider()
+            st.header('Step 2: Make it structured')
+            st.caption('Let\'s now convert the above generated contents into JSON')
+
+            # Streamlit multiple buttons work in a weird way!
+            # Click on any button, the page just reloads!
+            # Buttons are not "stateful"
+            # https://blog.streamlit.io/10-most-common-explanations-on-the-streamlit-forum/#1-buttons-aren%E2%80%99t-stateful
+            # Apparently, "nested button click" needs to be handled differently
+            # https://playground.streamlit.app/?q=triple-button
+
+            st.button(UI_BUTTONS[1], on_click=button_clicked, args=[1])
+
+            if st.session_state.clicked[1]:
+                progress_text = 'Converting...give it a moment'
+                progress_bar = st.progress(0, text=progress_text)
+
+                process_slides_contents(slides_content, progress_bar)
         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.')
 
-        # image = generate_image_from_text(summary)
-        # progress_bar.progress(100, text='Done!')
-        #
-        # st.image(image, caption=summary)
-        # st.info('Tip: Right-click on the image to save it')
     else:
         st.error('Not enough information provided! Please be little more descriptive :)')
 
 
+def process_slides_contents(text: str, progress_bar: st.progress):
+    """
+    Convert given content to JSON and display. Update the UI.
+
+    :param text: The contents generated for the slides
+    :param progress_bar: Progress bar for this step
+    """
+
+    print('JSON button clicked')
+    json_str = llm_helper.text_to_json(text)
+    print('=' * 20)
+    print(f'JSON:\n{json_str}')
+    print('=' * 20)
+    st.code(json_str, language='json')
+
+    progress_bar.progress(100, text='Done!')
+
+
+def button_clicked(button):
+    """
+    Function to update the value in session state.
+    """
+
+    st.session_state.clicked[button] = True
+
+
 def main():
+    # Initialize the key in session state to manage the nested buttons states
+    if 'clicked' not in st.session_state:
+        st.session_state.clicked = {0: False, 1: False, 2: False}
+
     build_ui()
 
 

global_config.py

@@ -12,11 +12,16 @@ class GlobalConfig:
     # Flan-T5
     # LLM_MODEL_NAME: str = 'google/flan-t5-xxl'
     LLM_MODEL_NAME = 'tiiuae/falcon-7b-instruct'
+    # LLM_MODEL_NAME = 'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-7b-v2'
+    # LLM_MODEL_NAME = 'garage-bAInd/Platypus2-70B-instruct'
     LLM_MODEL_TEMPERATURE: float = 0.5
-    LLM_MODEL_MIN_OUTPUT_LENGTH: int = 200
+    LLM_MODEL_MIN_OUTPUT_LENGTH: int = 50
     LLM_MODEL_MAX_OUTPUT_LENGTH: int = 2000
     LLM_MODEL_MAX_INPUT_LENGTH: int = 1000
 
     # # Stable Diffusion
     # DIFFUSION_MODEL_NAME: str = 'stabilityai/stable-diffusion-2-1'
     # DIFFUSION_NUM_INFERENCE_STEPS: int = 3
+
+    PRELOAD_DATA_FILE = 'examples/example_02.json'
+    SLIDES_TEMPLATE_FILE = 'langchain_templates/template_07.txt'

langchain_templates/template_07.txt

@@ -1,4 +1,4 @@
-You are a helpful, intelligent chatbot. Create the slides for a presentation on the given topic. Include main headings for each slide, detailed bullet points for each slide. Add relevant content to each slide.
+You are a helpful, intelligent chatbot. Create the slides for a presentation on the given topic. Include main headings for each slide, detailed bullet points for each slide. Add relevant content to each slide. Also, generate an engaging title for the presentation aimed for the target audience.
 
 
 Topic:
@@ -7,6 +7,3 @@ Topic:
 
 Target audience:
 ```{audience}```
-
-
-Finally, generate an engaging title for the presentation aimed for the target audience.

llm_agent.py

@@ -1,89 +0,0 @@
-from langchain import HuggingFaceHub, PromptTemplate
-
-from global_config import GlobalConfig
-
-
-llm = HuggingFaceHub(
-    repo_id=GlobalConfig.LLM_MODEL_NAME,
-    task='text-generation',
-    huggingfacehub_api_token=GlobalConfig.HUGGINGFACEHUB_API_TOKEN,
-    model_kwargs={
-        '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
-    }
-)
-print(llm)
-
-#
-# so that the speaker can deliver an engaging talk to the target audience. You use facts in your slides. When necessary,
-# you also look up online for further details
-
-template = '''
-You are an artificial intelligence assistant. 
-You are experienced in creating slides for a presentation on any given topic.
-Generate a slide deck based on the following information:
-
-
-Topic: 
-```{topic}```
-
-
-Target audience:
-```{audience}```
-
-
-Generate an engaging title for the presentation and output it as the first line.
-The next line should contain the speaker's name.
-Add a title and number to each slide on a separate line.
-Each slide should have a bulleted list of items to talk about.
-'''
-
-
-template2 = '''
-Act like a professional speaker and expert in PowerPoint: Create the outline for a PowerPoint presentation on 
-any given topic. Include main headings for each slide, detailed bullet points for each slide, 
-ideas for photos for each slide and an impactful intro and closing slide.
-
-
-Speaker's name:
-```{name}```
-
-
-Topic: 
-```{topic}```
-
-
-Target audience:
-```{audience}```
-
-
-Generate an engaging title for the presentation and output it in the first line.
-The next line should contain the speaker's name.
-Add a title and number to each slide on a separate line.
-'''
-
-# The contents of the slides should be in plain-text in the form of bulleted list items.
-
-with open('langchain_templates/template_07.txt', 'r') as in_file:
-    template_txt = in_file.read().strip()
-
-# prompt = PromptTemplate.from_template(template)
-prompt = PromptTemplate.from_template(template_txt)
-
-
-def generate_slides_content(name: str, topic: str, audience: str) -> str:
-    """
-    Generate the contents of slides for a presentation on a given topic.
-
-    :return: The summary
-    """
-
-    formatted_prompt = prompt.format(topic=topic, audience=audience)
-    print(formatted_prompt)
-
-    slides_content = llm(formatted_prompt)
-
-    return slides_content

llm_helper.py

@@ -0,0 +1,96 @@
+from langchain import HuggingFaceHub, PromptTemplate
+
+from global_config import GlobalConfig
+
+
+prompt = None
+
+
+def get_llm() -> HuggingFaceHub:
+    llm = HuggingFaceHub(
+        repo_id=GlobalConfig.LLM_MODEL_NAME,
+        task='text-generation',
+        huggingfacehub_api_token=GlobalConfig.HUGGINGFACEHUB_API_TOKEN,
+        model_kwargs={
+            '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
+        }
+    )
+    print(llm)
+
+    return llm
+
+
+def generate_slides_content(name: str, topic: str, audience: str) -> str:
+    """
+    Generate the outline/contents of slides for a presentation on a given topic.
+
+    :return: The content
+    """
+
+    global prompt
+
+    if not prompt:
+        with open(GlobalConfig.SLIDES_TEMPLATE_FILE, 'r') as in_file:
+            template_txt = in_file.read().strip()
+
+        prompt = PromptTemplate.from_template(template_txt)
+
+    formatted_prompt = prompt.format(topic=topic, audience=audience)
+    # print(formatted_prompt)
+
+    llm = get_llm()
+    slides_content = llm(formatted_prompt, verbose=True)
+
+    return slides_content
+
+
+def text_to_json(content: str) -> str:
+    """
+    Convert input text into structured JSON representation.
+
+    :param content: Input text
+    :return: JSON string
+    """
+
+    # f-string is not used in order to prevent interpreting the brackets
+    text = '''
+    Context:
+
+
+    '''
+    text += content
+    text += '''
+
+
+    Convert the above text into structured JSON output. The JSON structure should be something like this:
+    {
+        "presentation_title": "...",
+        "slides": [
+            {
+                "slide_number": "...",
+                "slide_heading": "...",
+                "slide_contents": [
+                    "...",
+                    "...",
+                ],
+            },
+            {
+                ...
+            },
+        ]
+    }
+    '''
+
+    llm = get_llm()
+    output = llm(text, verbose=True)
+    output = output.strip()
+
+    first_index = max(0, output.find('{'))
+    last_index = min(output.rfind('}'), len(output))
+    output = output[first_index: last_index + 1]
+
+    return output

requirements.txt

@@ -1,4 +1,4 @@
-python-dotenv[cli]
-langchain
+python-dotenv[cli]~=1.0.0
+langchain~=0.0.268
 huggingface_hub
-streamlit
+streamlit~=1.25.0