WhiteBoard SLLM (swe3032)

---

Project Paper & Slides

• Project Paper

• Project Slides

---

Structures

├── LICENSE
├── README.md
├── config.yaml
├── data
│   ├── collect_script
│   │   ├── combine_jsonl.py
│   │   ├── data_preparation.py
│   │   ├── preprocess.py
│   │   ├── setup.sh
│   │   └── transcript_extractor.py
│   ├── open_source
│   │   ├── huggingface_data_paths.txt
│   │   └── urls.txt
│   ├── youtube_urls
│   │   └── <category>_playlist.txt
│   ├── gov_indices<N>.npy
│   └── ytb_indices<N>.npy
├── experiment.ipynb
├── experiment.py
├── experiments
│   └── example
│       └── config.yaml
├── previous_project
│   ├── experiments
│   └── trainer_baseline.ipynb
└── utils
    ├── concat_functions.py
    ├── eval_similarity.py
    ├── segment_embedding.py
    ├── summarizer.py
    └── utils.py

Huggingface Organization

• Our team Organization Link

---

Dataset

• [./data]
  • [./data/opensource, youtube] : Open-source (Hugging Face) and YouTube metadata
  • [./data/collect_script/*] : Data extraction and preprocessing scripts
  • [./data/gov_indices<N>.npy]: Index sets sampled from the open-source dataset ccdv/govreport-summarization
  • [./data/ytb_indices<N>.npy]: Index sets sampled from our collected dataset WhiteboardLLM/Data

---

• actual dataset : Actual data is used after uploading to Hugging Face

  • Youtube GPT Summary datasets

  • Youtube Raw contents

  • Usage example: dataset = load_dataset(data_path)

---

Experiment

• [./config.yaml] : Main hyperparameters used for experiments
• [./experiment.py] : Experiment pipeline implementation (Python version)
• [./experiment.ipynb] : Experiment pipeline implementation (Notebook version)
• [./utils/*] : Function implementations for the current architecture
• [./experiments/*] : Experiment logs

---

Others (Previous Project)

• [./previous_project/*] : sLLM fine-tuning project

• trainer.ipynb : Training code used on Kaggle and Colab

• trained_models: Trained models are used after uploading to Hugging Face (see the Hugging Face repository)