README.md

metadata

title: Auto-PPT Generator
emoji: 📊
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false

Auto-PPT Generator (Hugging Face Space)

End-to-end pipeline: Long text → Summary → Sectioning → Bullets/Tables/Charts → PPTX export, with theme color and logo.

Run on a Space

1. Create a Gradio Space and upload all files.
2. (Optional) Settings → Variables & secrets: add HF_TOKEN if you will use the Inference API.
3. Click Run.

Models

• Local CPU (English): sshleifer/distilbart-cnn-12-6 (summarization)
• Local CPU (Japanese): sonoisa/t5-base-japanese (use text2text-generation path with 要約: prefix)
• Inference API: any instruct/summarization model you have access to (e.g., Qwen/Qwen2-7B-Instruct, elyza/ELYZA-japanese-Llama-2-7b-fast-instruct).

Input conventions

• Tables: Provide lines like 項目: 値 under a section to auto-build a 2-column table.
• Charts: Provide lines like ラベル: 123 (numbers) to auto-build a bar chart.
• Bullets: Lines starting with -, *, ・, or numbered lists are detected automatically.

Notes

• Slide numbers are approximated (python-pptx lacks true auto-numbering fields).
• For corporate fonts, pre-install them or post-process the PPTX if required.
• For very large texts, we truncate per-model token limits but the rule-based extractors remain robust.