README.md

---
title: README
emoji: 🌖
colorFrom: gray
colorTo: pink
sdk: static
pinned: false
---

The goal of the OpenLLM-Ro is to bring together the Romanian community that builds open Romanian models and to collect these models in a single place. 

We value:
- using public and open corpora
- open-source training and evaluation code.

In this organization, you can find RoLLM models, based on different underlying models and in different flavours (i.e., foundational, instruct, or chat variants):
- RoLlama2: Romanian models based on Llama2
- RoMistral: Romanian models based on Mistral
- RoLlama3: Romanian models based on Llama3
- RoLlama3.1: Romanian models based on Llama3.1
- RoGemma: Romanian models based on Gemma
- RoGemma2: Romanian models based on Gemma2
- RoLlava: Romanian models based on Llava
- RoQwen2-VL: Romanian models based on Qwen2-VL
- RoQwen2.5-VL: Romanian models based on Qwen2.5-VL
- RoQwen3-VL: Romanian models based on Qwen3-VL
- RoGemma3: Romanian models based on Gemma3


Furthermore, here you can find data used to train and evaluate LLMs & VLMs in Romanian. Currently, there are four data collections:
- Pretraining dataset: Romanian pretraining quality-filtered data
- SFT datasets: data used for supervised (instruction) finetuning
- Alignment datasets: data used mainly for Direct Preference Optimization (DPO)
- Evaluation datasets: data used for evaluating models in Romanian

See details in [https://arxiv.org/abs/2406.18266](https://arxiv.org/abs/2406.18266) and [https://arxiv.org/abs/2605.31401](https://arxiv.org/abs/2605.31401).

- 23-04-2025: we increased the datasets used for supervised finetuning with high-quality data generated using Magpie ([RoMagpie-Reasoning](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_magpie_reasoning) and [RoMagpie-Pro-MT](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_magpie_mt)), and greatly increase the size of the alignment dataset by adding high-quality datasets ([RoUltraFeedback](https://huggingface.co/datasets/OpenLLM-Ro/ro_dpo_ultrafeedback), [RoMagpie-DPO](https://huggingface.co/datasets/OpenLLM-Ro/ro_dpo_magpie), [RoArgillaMagpieUltra](https://huggingface.co/datasets/OpenLLM-Ro/ro_dpo_argilla_magpie) and [RoHelpSteer2](https://huggingface.co/datasets/OpenLLM-Ro/ro_dpo_helpsteer2))
- 28-11-2025: we release augmented pretraining data and quality classifier
- 05-06-2026: we release RoVLMs together with training and evaluation data
- 
We encourage the community to engage in discussions (to provide feedback, ask questions, or make improvement suggestions) in Hugging Face or GitHub. 

We will also organize physical meetings (announced in advance) to brainstorm ideas, roadmap, and other technical aspects.

Extra info: check also the work by the [Faur AI team](https://huggingface.co/faur-ai)