122 lines
3.8 KiB
Python
122 lines
3.8 KiB
Python
from typing import Optional
|
|
from google.genai.types import GenerateContentConfig
|
|
from openai.types.chat.chat_completion_chunk import ChoiceDelta
|
|
|
|
from utils.async_iterator import iterator_to_async
|
|
from utils.get_dynamic_models import get_presentation_outline_model_with_n_slides
|
|
from utils.llm_provider import (
|
|
get_google_llm_client,
|
|
get_large_model,
|
|
get_llm_client,
|
|
is_google_selected,
|
|
)
|
|
from pydantic import BaseModel
|
|
|
|
system_prompt = """
|
|
You are an expert presentation creator. Generate structured presentations based on user requirements and format them according to the specified JSON schema with markdown content.
|
|
|
|
## Core Requirements
|
|
|
|
### Input Processing
|
|
1. **Extract key information** from the user's prompt:
|
|
- Main topic/subject matter
|
|
- Required number of slides
|
|
- Target language for output
|
|
- Specific content requirements or focus areas
|
|
- Target audience (if specified)
|
|
- Presentation style or tone preferences
|
|
|
|
|
|
## Content Generation Guidelines
|
|
|
|
### Presentation Title
|
|
- Create a **concise, descriptive title** that captures the essence of the topic
|
|
- Use **plain text format** (no markdown formatting)
|
|
- Make it **engaging and professional**
|
|
- Ensure it reflects the main theme and target audience
|
|
|
|
### Slide Titles
|
|
- Generate **clear, specific titles** for each slide
|
|
- Use **plain text format** (no markdown, no "Slide 1", "Slide 2" prefixes)
|
|
- Make each title **descriptive and informative**
|
|
- Ensure titles create a **logical flow** through the presentation
|
|
- Keep titles **concise but meaningful**
|
|
|
|
|
|
## Special Considerations
|
|
|
|
### Slide Count Compliance
|
|
- Generate **exactly** the number of slides requested
|
|
- Distribute content **evenly** across slides
|
|
- Create **balanced information flow**
|
|
"""
|
|
|
|
|
|
def get_user_prompt(prompt: str, n_slides: int, language: str, content: str):
|
|
return f"""
|
|
**Input:**
|
|
- Prompt: {prompt}
|
|
- Output Language: {language}
|
|
- Number of Slides: {n_slides}
|
|
- Additional Information: {content}
|
|
"""
|
|
|
|
|
|
def get_prompt_template(prompt: str, n_slides: int, language: str, content: str):
|
|
return [
|
|
{
|
|
"role": "system",
|
|
"content": system_prompt,
|
|
},
|
|
{
|
|
"role": "user",
|
|
"content": get_user_prompt(prompt, n_slides, language, content),
|
|
},
|
|
]
|
|
|
|
|
|
def get_response_format(response_model: BaseModel):
|
|
return {
|
|
"type": "json_schema",
|
|
"json_schema": {
|
|
"name": "PresentationOutlineModel",
|
|
"schema": response_model.model_json_schema(),
|
|
},
|
|
}
|
|
|
|
|
|
async def generate_ppt_outline(
|
|
prompt: Optional[str],
|
|
n_slides: int,
|
|
language: Optional[str] = None,
|
|
content: Optional[str] = None,
|
|
):
|
|
model = get_large_model()
|
|
response_model = get_presentation_outline_model_with_n_slides(n_slides)
|
|
|
|
if not is_google_selected():
|
|
client = get_llm_client()
|
|
async for response in await client.chat.completions.create(
|
|
model=model,
|
|
messages=get_prompt_template(prompt, n_slides, language, content),
|
|
stream=True,
|
|
response_format=get_response_format(response_model),
|
|
):
|
|
delta: ChoiceDelta = response.choices[0].delta
|
|
if delta.content:
|
|
yield delta.content
|
|
|
|
else:
|
|
client = get_google_llm_client()
|
|
generate_stream = iterator_to_async(client.models.generate_content_stream)
|
|
async for event in generate_stream(
|
|
model=model,
|
|
contents=[get_user_prompt(prompt, n_slides, language, content)],
|
|
config=GenerateContentConfig(
|
|
system_instruction=system_prompt,
|
|
response_mime_type="application/json",
|
|
response_json_schema=response_model.model_json_schema(),
|
|
),
|
|
):
|
|
if event.text:
|
|
yield event.text
|