Quickstart

Get the Deltarune Complete Transcript Dataset loaded and queryable in Python in under five minutes.

1. Clone or download the repository

   Clone the repository from its source, or download and extract the archive.

   git clone https://github.com/ntvm/Deltarune-Complete-Transcript-Cleaned
   cd Deltarune-Complete-Transcript-Cleaned
   

   The repository contains two main data directories:

   data/
     chap1_dataset.jsonl
     chap2_dataset.jsonl
     chap3_dataset.jsonl
     chap4_dataset.jsonl
     chatml/
       deltarune_story_chatml.jsonl
   parquet/
     chap1_dataset.parquet
     chap2_dataset.parquet
     chap3_dataset.parquet
     chap4_dataset.parquet
     full_chapters_dataset.parquet
   
2. Install dependencies

   The dataset examples use pandas and pyarrow. Install them with your package manager of choice.

   pip

   pip install pandas pyarrow
   
3. Load a chapter from JSONL

   Each chapter is available as a JSONL file with one record per line. Use pandas.read_json with lines=True to load it into a DataFrame.

   load_jsonl.py

   import pandas as pd
   
   # Load a single chapter
   df = pd.read_json('data/chap1_dataset.jsonl', lines=True)
   
   print(df.shape)        # (1068, 3)
   print(df.columns.tolist())  # ['context', 'speaker', 'text']
   print(df.head())
   

   Example output:

                    context   speaker          text
   0  Scene: Device Contact  Narrator  ARE YOU THERE?
   1  Scene: Device Contact  Narrator  ARE WE CONNECTED?
   2  Scene: Device Contact  Narrator  ...
   
4. Load Parquet with pandas

   Parquet files load faster than JSONL for large queries. Use the per-chapter files or the combined file.

   load_parquet.py

   import pandas as pd
   
   # Load a single chapter as Parquet
   df_ch1 = pd.read_parquet('parquet/chap1_dataset.parquet')
   
   # Load all chapters at once
   df_all = pd.read_parquet('parquet/full_chapters_dataset.parquet')
   
   print(f"Total records: {len(df_all)}")
   print(df_all['speaker'].value_counts().head(10))
   

   Use full_chapters_dataset.parquet whenever you need to query across chapters — for example, to find every line spoken by a character regardless of chapter, or to analyze narrator patterns across the full game.

5. Filter by chapter, speaker, and scene

   All three fields — context, speaker, and text — support standard pandas filtering.

   filter_examples.py

   import pandas as pd
   
   df = pd.read_parquet('parquet/full_chapters_dataset.parquet')
   
   # Filter by speaker
   susie_lines = df[df['speaker'] == 'Susie']
   print(f"Susie has {len(susie_lines)} lines")
   
   # Filter narrator lines only
   narrator = df[df['speaker'] == 'Narrator']
   
   # Filter by scene (context field)
   cyber_world = df[df['context'] == 'Scene: Cyber World']
   
   # Combine filters: Ralsei lines in a specific scene
   ralsei_castle = df[
       (df['speaker'] == 'Ralsei') &
       (df['context'].str.contains('Castle', case=False))
   ]
   
   # Load a specific chapter by file
   ch2 = pd.read_parquet('parquet/chap2_dataset.parquet')
   ch2_player_choices = ch2[ch2['speaker'] == 'Player']
   
   print(ch2_player_choices['text'].tolist())
   

ChatML format

The data/chatml/deltarune_story_chatml.jsonl file contains the same transcript data pre-formatted as ChatML message sequences, suitable for supervised fine-tuning of instruction-following models.

Each record in this file uses the standard ChatML structure with three roles:

chatml example

{
  "messages": [
    {
      "role": "system",
      "content": "You are the Deltarune Script Archive. You provide exact, chronological scene transcripts including dialogues, actions, and pauses."
    },
    {
      "role": "user",
      "content": "Provide the transcript for Scene: Obj Krisroom.\nContext: This scene occurs after 'Device Contact' and before 'Obj Carcutscene'."
    },
    {
      "role": "assistant",
      "content": "Toriel: \"Kris...!\"\nToriel: \"Wake up!\"\nToriel: \"KRIS...!\"\nToriel: \"Kris, if you do not wake up, we will be late for school!\"\nToriel: \"I will wait outside for you, alright?\""
    }
  ]
}

Load it the same way as any other JSONL file:

load_chatml.py

import pandas as pd

chatml = pd.read_json('data/chatml/deltarune_story_chatml.jsonl', lines=True)
print(chatml.head())

See LLM fine-tuning for guidance on using this file with training frameworks.