BayanSimplify-v0.1

Model 1 of the Bayan project: AraT5v2 fine-tuned on SAMER (level 5 → 3).

What it does

The first Bayan model, kept for reference. On the SAMER test set it beats copying the input on the sentences a human editor changed (SARI 61.83 against 56.06, +5.77), but what it learned from SAMER is word substitution, not restructuring, and copying still scores higher over the whole test set. That finding shaped models 2 and 3.

Input: بسّط: (with the shadda) followed by the text. Decoding: greedy, 3-gram blocking, up to 256 tokens.

Use

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

repo = "Congi-libya/BayanSimplify-v0.1"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForSeq2SeqLM.from_pretrained(repo)

text = "وقد أدى التوسع العمراني السريع الذي شهدته المدينة خلال العقدين الماضيين إلى ازدحام مروري خانق في ساعات الذروة."
x = tok("بسّط: " + text, return_tensors="pt", max_length=256, truncation=True)
y = model.generate(**x, num_beams=1, no_repeat_ngram_size=3, max_length=256)
print(tok.decode(y[0], skip_special_tokens=True))

The BayanSimplify family

Model	What it is	Base model	Training data	In the Bayan app
BayanSimplify-v0.1	Model 1	AraT5v2-base-1024	SAMER, level 5 → 3	not shipped
BayanSimplify-v0.2	Model 2	AraT5v2-base-1024	22,733-pair mix (SAMER, Baseet, DAASI) with strength tags	"Earlier large model"
BayanSimplify-v0.2-Fast	Model 2, compact	AraBART	the same mix	"Fast model", the default download
BayanSimplify-v0.3	Model 3, our best	AraT5v2-base-1024	Bayan corpus v1, 14,975 rows	"Large model", four beams
BayanSimplify-ONNX	The int8 bundles the app downloads			v0.2-Fast, v0.2, v0.3

Licence and data

CC BY-NC 4.0 (non-commercial), in line with the training data's terms. SAMER is used under the CAMeL Lab's permission to fine-tune and share weights for non-commercial use; its text is not redistributed here.

About Bayan

Bayan (بيان) simplifies Arabic text for readers with dyslexia, entirely on an Android phone: select text in any app, choose «تبسيط» (Simplify), and a simpler version appears over the page. Built by Team Cogni for the Samsung Innovation Campus AI capstone, 2026. Every number on this card comes from the team's final report and BayanBench v2.0, the benchmark built for the task (meaning scored by Gemma 4 31B, checked against human raters, AUC 0.85).