You can deploy this model and start earning money today!

bbosseong/
byt5-multilexnorm-2026

Name: byt5-multilexnorm-2026
Availability: InStock

safetensors

Views

No views yet

MultiLexNorm 2026 Demo

This repository provides a demo for the MultiLexNorm shared task. It demonstrates how to download the dataset, run a simple baseline model (MFR), and evaluate normalization results.

Full code is available here.
The datasets will be available at development phase and final phase.
Example MFR submission outputs: outputs/submission_dev.zip and outputs/submission_full.zip

Set up the environment

bash

# Create an environment and install packages
python -m venv .venv
source .venv/bin/activate

pip install -r requirements.txt

Load data

python

from datasets import load_dataset

pub_data = load_dataset("weerayut/multilexnorm2026-pub")

# Select a language
lang = "en"
en_train = pub_data["train"].filter(lambda x: x["lang"] == lang)
en_val = pub_data["validation"].filter(lambda x: x["lang"] == lang)

Inference

python

import pandas as pd
from utils import counting, mfr

# Smoke test the baseline
counts = counting(en_train)
mfr(['bcause', 'u', 'r', 'funny'], counts)

# Inference
ds = pd.DataFrame(en_val)
ds['pred'] = ds['raw'].apply(lambda x: mfr(x, counts))

Evaluation

python

from utils import evaluate

evaluate(
    raw=ds['raw'].tolist(),    # list[list[str]]
    gold=ds['norm'].tolist(),  # list[list[str]]
    pred=ds['pred'].tolist()   # list[list[str]]
)

Output:

txt

Baseline acc.(LAI): 93.10
Accuracy:           97.37
ERR:                61.93

bbosseong/byt5-multilexnorm-2026

MultiLexNorm 2026 Demo

Set up the environment

Load data

Inference

Evaluation

bbosseong/
byt5-multilexnorm-2026