glitchlings 0.8.0

Monsters for your language games.

     .─') _                                       .─') _                  
    (  OO) )                                     ( OO ) )            
  ░██████  ░██ ░██   ░██               ░██        ░██ ░██                                 
 ░██   ░██ ░██       ░██                ░██        ░██                                     
░██        ░██ ░██░████████  ░███████   ░████████  ░██ ░██░████████   ░████████ ░███████  
░██  █████ ░██ ░██   ░██    ░██('─.░██ ░██    ░██ ░██ ░██░██    ░██ ░██.─')░██ ░██        
░██     ██ ░██ ░██   ░██    ░██( OO ) ╱░██    ░██ ░██ ░██░██    ░██ ░██(OO)░██ ░███████  
  ░██  ░███ ░██ ░██   ░██    ░██    ░██ ░██    ░██ ░██ ░██░██    ░██ ░██ o ░███      ░██ 
  ░█████░█ ░██ ░██   ░████   ░███████  ░██    ░██ ░██ ░██░██    ░██  ░█████░██ ░███████  
                                                                          ░██            
                                                                  ░███████             

                        Every language game breeds monsters.

Glitchlings are utilities for corrupting the text inputs to your language models in deterministic, linguistically principled ways.
Each embodies a different way that documents can be compromised in the wild.

If reinforcement learning environments are games, then Glitchlings are enemies to breathe new life into old challenges.

They do this by breaking surface patterns in the input while keeping the target output intact.

Some Glitchlings are petty nuisances. Some Glitchlings are eldritch horrors.
Together, they create truly nightmarish scenarios for your language models.

After all, what good is general intelligence if it can't handle a little chaos?

-The Curator

Motivation

If your model performs well on a particular task, but not when Glitchlings are present, it's a sign that it hasn't actually generalized to the problem.

Conversely, training a model to perform well in the presence of the types of perturbations introduced by Glitchlings should help it generalize better.

Quickstart

pip install -U glitchlings

The fastest way to get started is to ask my assistant, Auggie, to prepare a custom mix of glitchlings for you:

from glitchlings import Auggie, SAMPLE_TEXT

auggie = (
    Auggie(seed=404)
    .typo(rate=0.015)
    .confusable(rate=0.01)
    .homophone(rate=0.02)
)

print(auggie(SAMPLE_TEXT))

One morning, when Gregor Samsa woke from troubld dreams, he found himself transformed in his bed into a horible vermin. He layed on his armour-like back, and if he lifted his head a little he could see his brown belly, slightly domed and divided by arches into stiff sections. The bedding was hardly able to cover it and seemed ready to slide off any moment. His many legs, pitifully thin compared with the size of the rest of him, waved about helplessly as he looked.

You're more than welcome to summon them directly, if you're feeling brave:

from glitchlings import Gaggle, SAMPLE_TEXT, Typogre, Mim1c, Ekkokin

gaggle = Gaggle(
    [
        Typogre(rate=0.015),
        Mim1c(rate=0.01),
        Ekkokin(rate=0.02),
    ],
    seed=404
)

Consult the Glitchlings Usage Guide for end-to-end instructions spanning the Python API, CLI, HuggingFace, PyTorch, and Prime Intellect integrations, and the compiled Rust pipeline (always enabled).

Your First Battle

Summon your chosen Glitchling (or a few, if ya nasty) and call it on your text or slot it into Dataset.map(...), supplying a seed if desired. Glitchlings are standard Python classes, so you can instantiate them with whatever parameters fit your scenario:

from glitchlings import Gaggle, Typogre, Mim1c

custom_typogre = Typogre(rate=0.1)
selective_mimic = Mim1c(rate=0.05, classes=["LATIN", "GREEK"])

gaggle = Gaggle([custom_typogre, selective_mimic], seed=99)
print(gaggle("Summoned heroes do not fear the glitch."))

Calling a Glitchling on a str transparently calls .corrupt(str, ...) -> str. This means that as long as your glitchlings get along logically, they play nicely with one another.

When summoned as or gathered into a Gaggle, the Glitchlings will automatically order themselves into attack waves, based on the scope of the change they make:

1. Document
2. Paragraph
3. Sentence
4. Word
5. Character

They're horrible little gremlins, but they're not unreasonable.

Command-Line Interface (CLI)

Keyboard warriors can challenge them directly via the glitchlings command (see the generated CLI reference in docs/cli.md for the full contract):

# Discover which glitchlings are currently on the loose.
glitchlings --list
 
# Review the full CLI contract.
glitchlings --help
 
# Run Typogre against the contents of a file and inspect the diff.
glitchlings -g typogre --file documents/report.txt --diff

# Configure glitchlings inline by passing keyword arguments.
glitchlings -g "Typogre(rate=0.05)" "Ghouls just wanna have fun"

# Pipe text straight into the CLI for an on-the-fly corruption.
echo "Beware LLM-written flavor-text" | glitchlings -g mim1c

Attack Configurations

Attack configurations live in plain YAML files so you can version-control experiments without touching code:

# Load a roster from a YAML attack configuration.
glitchlings --config experiments/chaos.yaml "Let slips the glitchlings of war"

# experiments/chaos.yaml
seed: 31337
glitchlings:
  - name: Typogre
    rate: 0.04
  - "Rushmore(rate=0.12, unweighted=True)"
  - name: Zeedub
    parameters:
      rate: 0.02
      characters: ["\u200b", "\u2060"]

Attack on Token

Looking to compare before/after corruption with metrics and stable seeds? Reach for the Attack helper, which bundles tokenization, metrics, and transcript batching into a single utility.

Development

Follow the development setup guide for editable installs, automated tests, and tips on enabling the Rust pipeline while you hack on new glitchlings.

Starter 'lings

For maintainability reasons, all Glitchling have consented to be given nicknames once they're in your care. See the Monster Manual for a complete bestiary.

Typogre

What a nice word, would be a shame if something happened to it.

Fatfinger. Typogre introduces character-level errors (duplicating, dropping, adding, or swapping) based on the layout of a keyboard (QWERTY by default, with Dvorak and Colemak variants built-in).

Mim1c

Wait, was that...?

Confusion. Mim1c replaces non-space characters with Unicode Confusables, characters that are distinct but would not usually confuse a human reader.

Hokey

She's soooooo coooool!

Passionista. Hokey gets a little excited and streeeeetches words for emphasis.

Apocryphal Glitchling contributed by Chloé Nunes

Scannequin

How can a computer need reading glasses?

OCR Artifacts. Scannequin mimics optical character recognition errors by swapping visually similar character sequences (like rn↔m, cl↔d, O↔0, l/I/1).

Zeedub

Watch your step around here.

Invisible Ink. Zeedub slips zero-width codepoints between non-space character pairs, forcing models to reason about text whose visible form masks hidden glyphs.

Ekkokin

Did you hear what I heard?

Echo Chamber. Ekkokin swaps words with curated homophones so the text still sounds right while the spelling drifts. Groups are normalised to prevent duplicates and casing is preserved when substitutions fire.

Jargoyle

Uh oh. The worst person you know just bought a thesaurus.

Sesquipedalianism. Jargoyle insufferably replaces words from selected parts of speech with synonyms at random, without regard for connotational or denotational differences.

Rushmore

I accidentally an entire word.

Tactical Scrambler. Rushmore randomly drops, duplicates, or swaps words in the text to simulate hasty writing, editing mistakes, or transmission errors.

Redactyl

Oops, that was my black highlighter.

FOIA Reply. Redactyl obscures random words in your document like an NSA analyst with a bad sense of humor.

Apocrypha

Cave paintings and oral tradition contain many depictions of strange, otherworldly Glitchlings.
These Apocryphal Glitchling are said to possess unique abilities or behaviors.
If you encounter one of these elusive beings, please document your findings and share them with The Curator.

Ensuring Reproducible Corruption

Every Glitchling should own its own independent random.Random instance. That means:

• No random.seed(...) calls touch Python's global RNG.
• Supplying a seed when you construct a Glitchling (or when you summon(...)) makes its behavior reproducible.
• Re-running a Gaggle with the same master seed and the same input text (and same external data!) yields identical corruption output.
• Corruption functions are written to accept an rng parameter internally so that all randomness is centralized and testable.

At Wits' End?

If you're trying to add a new glitchling and can't seem to make it deterministic, here are some places to look for determinism-breaking code:

1. Search for any direct calls to random.choice, random.shuffle, or set(...) ordering without going through the provided rng.
2. Ensure you sort collections before shuffling or sampling.
3. Make sure indices are chosen from a stable reference (e.g., original text) when applying length‑changing edits.
4. Make sure there are enough sort keys to maintain stability.