AfroXLMR-large-76L-Injongo-slot

McGill-NLP/AfroXLMR-large-76L-Injongo-slot

AfroXLMR-large-76L-Injongo-slot is a token classification model from McGill-NLP. Use it when you need labels on individual words, such as names. It is set up for transformers. The card lists the license as cc-by-4.0.

- eng: English - amh: Amharic - ewe: Ewe - hau: Hausa - ibo: Igbo - kin: Kinyarwanda - lin: Lingala - lug: Luganda - orm: Oromo - sna: Shona - sot: Sesotho - swa: Swahili - twi: Twi - wol: Wolof - xho: Xhosa - yor: Yo…

Downloads · 30 days

31% of all-time downloads

All-time downloads

Public

Parameters

559M

2.3 GB on disk

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors2.2 GB · 99%

At a glance

Task: Token Classification
Library: transformers
License: cc-by-4.0
Model type: xlm-roberta
Access: Public
Created: May 24, 2025
Updated: May 25, 2025
SHA: ead9d911

Try a prompt

Base models

Davlan/afro-xlmr-large-76L

Datasets

masakhane/InjongoIntent

Model card

From the Hugging Face model README

INJONGO: A Multicultural Intent Detection and Slot-filling Dataset for 16 African Languages

Evaluation Comparison

Intent Detection (Accuracy)

In-language training

Model	eng	amh	ewe	hau	ibo	kin	lin	lug	orm	sna	sot	swa	twi	wol	xho	yor	zul	AVG
mT5-Large	80.5	91.5	77.3	94.6	92.9	83.7	91.3	83.3	73.3	92.6	80.2	95.8	85.3	91.6	95.8	90.9	82.4	87.7±4.1
AfriTeVa V2 (T5)	81.6	93.2	84.4	98.9	95.7	87.8	91.6	86.8	86.6	94.6	85.7	96.8	87.1	94.0	97.3	97.0	89.2	91.7±2.7
NLLB LLM2Vec	88.4	94.2	87.8	98.3	96.8	89.2	95.2	93.2	86.2	96.1	87.3	97.4	93.5	95.6	97.5	97.3	89.1	93.4±2.3
XLM-RoBERTa	83.5	92.9	77.9	96.0	88.8	69.6	90.5	78.9	75.0	83.8	76.0	96.7	79.5	90.2	89.6	92.6	74.7	84.5±4.9
AfriBERTa V2	74.2	91.2	78.3	98.2	93.8	83.1	91.0	83.8	78.8	89.5	81.9	96.0	83.2	92.3	94.4	95.0	86.7	88.6±3.5
AfroXLMR	84.1	95.3	84.6	98.3	96.0	88.2	93.3	85.2	88.3	95.3	85.5	97.8	88.8	95.8	97.3	96.1	89.0	92.2±3.0
AfroXLMR 76L	84.5	95.5	90.4	98.7	96.3	89.4	94.6	91.3	88.3	95.1	86.8	98.1	93.6	96.2	96.9	97.7	89.8	93.7±2.1

Multi-lingual training

Model	eng	amh	ewe	hau	ibo	kin	lin	lug	orm	sna	sot	swa	twi	wol	xho	yor	zul	AVG
AfroXLMR-large-76L-Injongo-intent	89.0	96.0	92.6	99.2	96.6	87.7	95.9	92.3	92.9	96.5	87.6	97.8	94.2	97.1	97.3	97.9	89.2	94.4±2.0

Slot Filling (F1)

In-language training

Model	eng	amh	ewe	hau	ibo	kin	lin	lug	orm	sna	sot	swa	twi	wol	xho	yor	zul	AVG
mT5-Large	73.7	80.9	71.6	89.4	80.5	74.2	82.6	78.9	72.1	81.1	74.7	88.1	79.0	76.9	88.4	78.9	68.3	79.1±3.7
AfriTeVa V2 (T5)	73.6	80.9	74.5	93.8	79.9	76.6	87.1	85.2	79.0	82.1	77.5	88.9	84.0	79.0	90.0	87.2	71.2	82.3±3.3
NLLB LLM2Vec	74.6	82.4	80.5	93.6	78.1	70.1	84.8	86.6	80.8	81.4	74.8	85.7	85.7	78.3	88.0	85.0	78.3	82.1±3.1
XLM-RoBERTa	77.9	84.8	79.9	93.9	76.6	69.3	86.3	83.8	83.8	79.3	71.7	88.7	84.2	79.3	89.1	83.9	79.4	82.1±3.5
AfriBERTa V2	70.7	82.2	77.9	93.7	78.3	73.8	84.4	84.1	81.0	81.8	73.5	87.6	81.9	78.3	88.5	86.2	79.6	82.1±2.9
AfroXLMR	79.0	86.2	81.6	95.1	82.0	76.3	87.1	88.5	84.9	84.9	77.5	90.2	85.5	81.7	91.1	87.3	82.5	85.2±2.7
AfroXLMR 76L	78.7	86.3	84.5	94.3	81.9	76.7	88.0	88.8	85.5	84.9	77.4	90.2	89.8	81.3	90.5	88.1	81.3	85.6±2.7

Multi-lingual training

Model	eng	amh	ewe	hau	ibo	kin	lin	lug	orm	sna	sot	swa	twi	wol	xho	yor	zul	AVG
AfroXLMR-large-76L-Injongo-slot	82.4	88.2	87.0	96.3	84.0	79.3	90.3	89.2	87.2	86.1	80.4	90.5	90.3	83.3	91.8	90.2	83.3	87.3±2.4

Language Codes

eng: English
amh: Amharic
ewe: Ewe
hau: Hausa
ibo: Igbo
kin: Kinyarwanda
lin: Lingala
lug: Luganda
orm: Oromo
sna: Shona
sot: Sesotho
swa: Swahili
twi: Twi
wol: Wolof
xho: Xhosa
yor: Yoruba
zul: Zulu

Notes

Bold values indicate the best performing scores in each category
The highlighted models (AfroXLMR 76L) show the top overall performance
Multi-lingual training generally outperforms in-language training
Standard deviations are reported alongside average scores
AVG doest not include english results.

Citation

@misc{yu2025injongo,
      title={INJONGO: A Multicultural Intent Detection and Slot-filling Dataset for 16 African Languages}, 
      author={Hao Yu and Jesujoba O. Alabi and Andiswa Bukula and Jian Yun Zhuang and En-Shiun Annie Lee and Tadesse Kebede Guge and Israel Abebe Azime and Happy Buzaaba and Blessing Kudzaishe Sibanda and Godson K. Kalipe and Jonathan Mukiibi and Salomon Kabongo Kabenamualu and Mmasibidi Setaka and Lolwethu Ndolela and Nkiruka Odu and Rooweither Mabuya and Shamsuddeen Hassan Muhammad and Salomey Osei and Sokhar Samb and Juliet W. Murage and Dietrich Klakow and David Ifeoluwa Adelani},
      year={2025},
      eprint={2502.09814},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2502.09814}, 
}

@misc{adelani2023sib200,
      title={SIB-200: A Simple, Inclusive, and Big Evaluation Dataset for Topic Classification in 200+ Languages and Dialects}, 
      author={David Ifeoluwa Adelani and Hannah Liu and Xiaoyu Shen and Nikita Vassilyev and Jesujoba O. Alabi and Yanke Mao and Haonan Gao and Annie En-Shiun Lee},
      year={2023},
      eprint={2309.07445},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Task: Token Classification
Library: transformers
Type: xlm-roberta
License: cc-by-4.0
Languages: en, am, ee, ha, ig, rw
Created: May 24, 2025
Updated: May 25, 2025