---
library_name: transformers
license: other
license_name: mixed
pipeline_tag: text-generation
language:
- ar
- es
- cs
- de
- id
- tl
- ja
base_model:
- allenai/Olmo-3-1025-7B
- google/gemma-3-4b-pt
datasets:
- ljvmiranda921/PolyglotTeachers-SFT-Synth
tags:
- multilingual
- synthetic
- sft
---


<div style="display: flex; align-items: center; gap: 20px;">
  <img alt="Logo for UCam" src="cambridge_logo.png" style="height: 80px; width: auto;">
  <img alt="Logo for LTL" src="ltl_logo2.svg" style="height: 80px; width: auto;">
</div>

**Website:** [ljvmiranda921.github.io/polyglot-teachers/](https://ljvmiranda921.github.io/polyglot-teachers/)


# Multilingual Instruct Models (Polyglot Teachers)

These are per-language models supervised fine-tuned on the synthetic data
generated in the [Polyglot Teachers](https://huggingface.co/papers/2604.11290)
project (see [ljvmiranda921/PolyglotTeachers-SFT-Synth](https://huggingface.co/datasets/ljvmiranda921/PolyglotTeachers-SFT-Synth)).

Load a specific model by passing the branch as the `revision`:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "ljvmiranda921/Polyglot-SFT-Multilingual"
branch = "Polyglot-OLMo3-7B-SFT-ar"  # pick any branch below
model = AutoModelForCausalLM.from_pretrained(repo, revision=branch)
tokenizer = AutoTokenizer.from_pretrained(repo, revision=branch)
```

## Branches

| Branch | Description |
| --- | --- |
| `Polyglot-Gemma3-4B-SFT-ar` | Gemma-3 4B SFT — Arabic |
| `Polyglot-Gemma3-4B-SFT-de` | Gemma-3 4B SFT — German |
| `Polyglot-Gemma3-4B-SFT-id` | Gemma-3 4B SFT — Indonesian |
| `Polyglot-Gemma3-4B-SFT-tl` | Gemma-3 4B SFT — Tagalog |
| `Polyglot-OLMo3-7B-SFT-ar` | OLMo-3 7B SFT — Arabic |
| `Polyglot-OLMo3-7B-SFT-cs` | OLMo-3 7B SFT — Czech |
| `Polyglot-OLMo3-7B-SFT-de` | OLMo-3 7B SFT — German |
| `Polyglot-OLMo3-7B-SFT-es` | OLMo-3 7B SFT — Spanish |
| `Polyglot-OLMo3-7B-SFT-id` | OLMo-3 7B SFT — Indonesian |
| `Polyglot-OLMo3-7B-SFT-ja` | OLMo-3 7B SFT — Japanese |


## Licensing

This repo holds models under different licenses; each branch follows its base
model's license:

- `Polyglot-OLMo3-7B-SFT-*` (base [allenai/Olmo-3-1025-7B](https://huggingface.co/allenai/Olmo-3-1025-7B)) — Apache-2.0
- `Polyglot-Gemma3-4B-SFT-*` (base [google/gemma-3-4b-pt](https://huggingface.co/google/gemma-3-4b-pt)) — [Gemma license](https://ai.google.dev/gemma/terms)

## Acknowledgements

LJVM and AK acknowledge the support of the UKRI Frontier Grant EP/Y031350/1 ([EQUATE](https://gtr.ukri.org/projects?ref=EP%2FY031350%2F1)).
This work was performed using joint resources provided by the [Cambridge Service for Data Driven Discovery (CSD3)](https://hpc.cam.ac.uk/high-performance-computing) EP/T022159/1 and the [Isambard AI National AI Research Resource (AIRR)](https://www.bristol.ac.uk/research/centres/bristol-supercomputing/#isambard-ai) ST/AIRR/I-A-I/1023, and the Microsoft Research Grant.
LJVM would also like to thank Songbo Hu, Chen Cecilia Liu, Millicent Ochieng, and Felermino Ali for helpful and productive discussions on the project.

## Citation

```bibtex
@inproceedings{miranda-etal-2026-polyglot,
    title = "Polyglot Teachers: Evaluating Language Models for Multilingual Synthetic Data Generation",
    author = "Miranda, Lester James Validad  and
      Vuli{\'c}, Ivan  and
      Korhonen, Anna",
    editor = "Lee, Hung-Yi  and
      Dev, Sunipa  and
      Li, Jessy  and
      Elliott, Desmond",
    booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
    month = oct,
    year = "2026",
    address = "Budapest, Hungary",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.emnlp-main.440/",
    pages = "10116--10144",
    ISBN = "979-8-89176-459-0",
    abstract = "Synthesizing supervised finetuning (SFT) data from language models (LMs) to teach smaller models multilingual tasks has become increasingly common. However, teacher model selection is often ad hoc, typically defaulting to the largest available option, even though such models may have significant capability gaps in non-English languages. This practice can result in poor-quality synthetic data and suboptimal student downstream performance. In this work, we systematically characterize what makes an effective multilingual teacher. We combine intrinsic measures of data quality with extrinsic student model performance in a metric we call Polyglot Score. We evaluate 10 LMs across 6 typologically diverse languages, generating over 1.4M SFT examples and training 240 student models. Our analyses reveal that model scale alone does not significantly predict teacher effectiveness: the most effective teachers we identify are consistently smaller than the largest models evaluated, and their ranking is stable across student base model families. Instead, data qualities such as prompt diversity, length, and response fluency capture 93.3{\%} of the variance in intrinsic data quality and predict student performance. Finally, we provide practical recommendations, including matching the model families of teacher-student pairs and generating responses to existing prompts or translating them from English, which can yield improvements for less-resourced languages. We hope that our work advances data-centric research in multilingual synthetic data and LM development."
}
```