Update prompts to generate image keywords in the JSON output

global_config.py

@@ -1,3 +1,6 @@
+"""
+A set of configurations used by the app.
+"""
 import logging
 import os
 
@@ -10,10 +13,14 @@ load_dotenv()
 
 @dataclass(frozen=True)
 class GlobalConfig:
-    HF_LLM_MODEL_NAME = 'mistralai/Mistral-7B-Instruct-v0.2'
+    """
+    A data class holding the configurations.
+    """
+
+    HF_LLM_MODEL_NAME = 'mistralai/Mistral-Nemo-Instruct-2407'
     LLM_MODEL_TEMPERATURE: float = 0.2
-    LLM_MODEL_MIN_OUTPUT_LENGTH: int = 50
-    LLM_MODEL_MAX_OUTPUT_LENGTH: int = 4096
+    LLM_MODEL_MIN_OUTPUT_LENGTH: int = 100
+    LLM_MODEL_MAX_OUTPUT_LENGTH: int = 4 * 4096
     LLM_MODEL_MAX_INPUT_LENGTH: int = 750
 
     HUGGINGFACEHUB_API_TOKEN = os.environ.get('HUGGINGFACEHUB_API_TOKEN', '')
@@ -25,8 +32,8 @@ class GlobalConfig:
     PRELOAD_DATA_FILE = 'examples/example_02.json'
     SLIDES_TEMPLATE_FILE = 'langchain_templates/template_combined.txt'
     # JSON_TEMPLATE_FILE = 'langchain_templates/text_to_json_template_02.txt'
-    INITIAL_PROMPT_TEMPLATE = 'langchain_templates/chat_prompts/initial_template_v3_two_cols.txt'
-    REFINEMENT_PROMPT_TEMPLATE = 'langchain_templates/chat_prompts/refinement_template_v3_two_cols.txt'
+    INITIAL_PROMPT_TEMPLATE = 'langchain_templates/chat_prompts/initial_template_v4_two_cols_img.txt'
+    REFINEMENT_PROMPT_TEMPLATE = 'langchain_templates/chat_prompts/refinement_template_v4_two_cols_img.txt'
 
     PPTX_TEMPLATE_FILES = {
         'Basic': {

langchain_templates/chat_prompts/initial_template_v4_two_cols_img.txt

@@ -0,0 +1,86 @@
+You are a helpful, intelligent chatbot. You are experienced with PowerPoint.
+
+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.
+The content of each slide should be VERBOSE, DESCRIPTIVE, and very DETAILED.
+If relevant, add one or two EXAMPLES to illustrate the concept.
+For two or three important slides, generate the key message that those slides convey.
+Identify if a slide describes a step-by-step/sequential process, then begin the bullet points with a special marker >>. Limit this to max two or three slides.
+Also, add at least one slide with a double column layout by generating appropriate content based on the description in the JSON schema provided below.
+In addition, for each slide, add image keywords based on the content of the respective slides.
+These keywords will be later used to search for images from the Web relevant to the slide content.
+ALWAYS add a concluding slide at the end, containing a list of the key takeaways and an optional call-to-action if relevant to the context.
+Unless explicitly instructed, create 10 TO 12 SLIDES in total.
+
+
+### Topic:
+{question}
+
+
+The output must be only a valid and syntactically correct JSON adhering to the following schema:
+{{
+    "title": "Presentation Title",
+    "slides": [
+        {{
+            "heading": "Heading for the First Slide",
+            "bullet_points": [
+                "First bullet point",
+                [
+                    "Sub-bullet point 1",
+                    "Sub-bullet point 2"
+                ],
+                "Second bullet point"
+            ],
+            "key_message": "",
+            "img_keywords": "a few keywords"
+        }},
+        {{
+            "heading": "Heading for the Second Slide",
+            "bullet_points": [
+                "First bullet point",
+                "Second bullet item",
+                "Third bullet point"
+            ],
+            "key_message": "The key message conveyed in this slide",
+            "img_keywords": "some keywords for this slide"
+        }},
+        {{
+            "heading": "A slide that describes a step-by-step/sequential process",
+            "bullet_points": [
+                ">> The first step of the process (begins with special marker >>)",
+                ">> A second step (begins with >>)",
+                ">> Third step",
+            ],
+            "key_message": "",
+            "img_keywords": ""
+        }},
+        {{
+            "heading": "A slide with a double column layout (useful for side-by-side comparison/contrasting of two related concepts, e.g., pros & cons, advantages & risks, old approach vs. modern approach, and so on)",
+            "bullet_points": [
+                {{
+                    "heading": "Heading of the left column",
+                    "bullet_points": [
+                        "First bullet point",
+                        "Second bullet item",
+                        "Third bullet point"
+                    ]
+                }},
+                {{
+                    "heading": "Heading of the right column",
+                    "bullet_points": [
+                        "First bullet point",
+                        "Second bullet item",
+                        "Third bullet point"
+                    ]
+                }}
+            ],
+            "key_message": "",
+            "img_keywords": ""
+        }}
+    ]
+}}
+
+
+### Output:
+```json

langchain_templates/chat_prompts/refinement_template_v4_two_cols_img.txt

@@ -0,0 +1,92 @@
+You are a helpful, intelligent chatbot. You are experienced with PowerPoint.
+
+A list of user instructions is provided below in sequential order -- from the oldest to the latest.
+The previously generated content of the slide deck in JSON format is also provided.
+Follow the instructions to revise the content of the previously generated slides of the presentation on the given topic.
+Include main headings for each slide, detailed bullet points for each slide.
+Add relevant content to each slide.
+The content of each slide should be VERBOSE, DESCRIPTIVE, and very DETAILED.
+If relevant, add one or two EXAMPLES to illustrate the concept.
+For two or three important slides, generate the key message that those slides convey.
+Identify if a slide describes a step-by-step/sequential process, then begin the bullet points with a special marker >>. Limit this to max two or three slides.
+Also, add at least one slide with a double column layout by generating appropriate content based on the description in the JSON schema provided below.
+In addition, for each slide, add image keywords based on the content of the respective slides.
+These keywords will be later used to search for images from the Web relevant to the slide content.
+ALWAYS add a concluding slide at the end, containing a list of the key takeaways and an optional call-to-action if relevant to the context.
+Unless explicitly instructed, create 10 TO 12 SLIDES in total.
+
+
+### List of instructions:
+{instructions}
+
+
+### Previously generated slide deck content as JSON:
+{previous_content}
+
+
+The output must be only a valid and syntactically correct JSON adhering to the following schema:
+{{
+    "title": "Presentation Title",
+    "slides": [
+        {{
+            "heading": "Heading for the First Slide",
+            "bullet_points": [
+                "First bullet point",
+                [
+                    "Sub-bullet point 1",
+                    "Sub-bullet point 2"
+                ],
+                "Second bullet point"
+            ],
+            "key_message": "",
+            "img_keywords": "a few keywords"
+        }},
+        {{
+            "heading": "Heading for the Second Slide",
+            "bullet_points": [
+                "First bullet point",
+                "Second bullet item",
+                "Third bullet point"
+            ],
+            "key_message": "The key message conveyed in this slide",
+            "img_keywords": "some keywords for this slide"
+        }},
+        {{
+            "heading": "A slide that describes a step-by-step/sequential process",
+            "bullet_points": [
+                ">> The first step of the process (begins with special marker >>)",
+                ">> A second step (begins with >>)",
+                ">> Third step",
+            ],
+            "key_message": "",
+            "img_keywords": ""
+        }},
+        {{
+            "heading": "A slide with a double column layout (useful for side-by-side comparison/contrasting of two related concepts, e.g., pros & cons, advantages & risks, old approach vs. modern approach, and so on)",
+            "bullet_points": [
+                {{
+                    "heading": "Heading of the left column",
+                    "bullet_points": [
+                        "First bullet point",
+                        "Second bullet item",
+                        "Third bullet point"
+                    ]
+                }},
+                {{
+                    "heading": "Heading of the right column",
+                    "bullet_points": [
+                        "First bullet point",
+                        "Second bullet item",
+                        "Third bullet point"
+                    ]
+                }}
+            ],
+            "key_message": "",
+            "img_keywords": ""
+        }}
+    ]
+}}
+
+
+### Output:
+```json