Modelpublic

OpenLLM-Ro/RoQwen3-VL-2B-Instruct

sourceHugging Facecc-by-nc-4.0updated 4mo agoView on Hugging Face

0likes27downloads

Model Card

Model Card for RoQwen3-VL-2B-Instruct

RoQwen3-VL-2B-Instruct is a Romanian-adapted vision-language model built on top of Qwen/Qwen3-VL-2B-Instruct. It was produced by continued supervised instruction tuning of the base Qwen3-VL checkpoint on a Romanian multimodal SFT mixture covering general instruction following (LLaVA mix), captioning (Pixmo-Cap, Flickr30k-Cap), visual question answering (Pixmo-AA, Pixmo-Cap-QA, Flickr30k-QA), document and chart understanding (CoSyn, FinePDFs), and visual grounding (Pixmo-Points, Pixmo-Count). The model is intended for research on Romanian VLM capabilities.

Model Details

Model Description

—Developed by: OpenLLM-Ro
—Language(s): Romanian
—License: cc-by-nc-4.0
—Finetuned from model: Qwen/Qwen3-VL-2B-Instruct
—Trained using:
—OpenLLM-Ro/ro_sft_laion
—OpenLLM-Ro/ro_sft_pixmo_cap
—OpenLLM-Ro/ro_sft_flickr30k_cap
—OpenLLM-Ro/ro_sft_llava_mix
—OpenLLM-Ro/ro_sft_pixmo_aa
—OpenLLM-Ro/ro_sft_pixmo_cap_qa
—OpenLLM-Ro/ro_sft_flickr30k_qa
—OpenLLM-Ro/ro_sft_cosyn
—OpenLLM-Ro/ro_sft_finepdfs
—OpenLLM-Ro/ro_sft_pixmo_points
—OpenLLM-Ro/ro_sft_pixmo_count

Model Sources

—Repository: https://github.com/OpenLLM-Ro/LLaMA-Factory
—Paper: https://arxiv.org/abs/2605.31401

Intended Use

Intended Use Cases

RoQwen3-VL-2B-Instruct is intended for research use on Romanian vision-language tasks — captioning, visual question answering, cultural understanding, OCR / document understanding, and visual grounding — and as a starting point for further Romanian VLM adaptation.

Out-of-Scope Use

Use in any manner that violates applicable laws or regulations (including trade-compliance laws), the project's license, or use in languages other than Romanian.

How to Get Started with the Model

python

import torch
from PIL import Image
from transformers import AutoProcessor, Qwen3VLForConditionalGeneration

model = Qwen3VLForConditionalGeneration.from_pretrained(
    "OpenLLM-Ro/RoQwen3-VL-2B-Instruct",
    torch_dtype=torch.bfloat16,
    device_map="auto",
).eval()
processor = AutoProcessor.from_pretrained("OpenLLM-Ro/RoQwen3-VL-2B-Instruct")

image = Image.open("example.jpg").convert("RGB")
question = "Descrie imaginea în detaliu."

messages = [
    {"role": "user", "content": [
        {"type": "image", "image": image},
        {"type": "text", "text": question},
    ]},
]
inputs = processor.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device, dtype=torch.bfloat16)

with torch.inference_mode():
    outputs = model.generate(**inputs, max_new_tokens=256, do_sample=False)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))

Benchmarks

All benchmarks below are evaluated in Romanian. Per-benchmark winners are shown in bold. Micro is the mean over individual benchmarks; Macro is the mean over capability groups.

Aggregate

Model	Micro avg.	Macro avg.
Qwen3-VL-2B-Instruct	51.51	51.31
RoQwen3-VL-2B-Instruct	63.36	62.65

General Understanding

Model	MMBench	MMStar	SeedBench2
Qwen3-VL-2B-Instruct	62.69	45.92	63.38
RoQwen3-VL-2B-Instruct	71.90	50.73	69.29

Knowledge & Reasoning

Model	MMMU	MME
Qwen3-VL-2B-Instruct	38.33	61.59
RoQwen3-VL-2B-Instruct	40.22	62.19

Cultural

Model	CVQA	ALM-Bench	RoMemes	RoCultVLM
Qwen3-VL-2B-Instruct	57.95	48.72	46.68	50.31
RoQwen3-VL-2B-Instruct	61.92	60.97	36.71	54.00

Generation & Open-ended

Model	RoFlickr30k-Caption	RoFlickr30k-QA	LLaVA-Wild	AyaVisionBench	m-WildVision
Qwen3-VL-2B-Instruct	70.09	30.59	29.89	43.04	44.76
RoQwen3-VL-2B-Instruct	83.80	85.70	50.40	55.33	60.08

OCR & Documents

Model	RoCosyn	RoFinepdfs	RoMemes OCR
Qwen3-VL-2B-Instruct	48.63	78.62	91.04
RoQwen3-VL-2B-Instruct	64.07	86.85	89.54

Grounding

Model	PixmoCount	PixmoPoints
Qwen3-VL-2B-Instruct	56.36	10.09
RoQwen3-VL-2B-Instruct	65.28	54.89

Citation

bibtex

@misc{masala2026intelegi,
      title={``\^{I}n\c{t}elegi Rom\^{a}ne\c{s}te?'' A Recipe for Romanian Vision-Language Models},
      author={Mihai Masala and Marius Leordeanu and Mihai Dascalu and Traian Rebedea},
      year={2026},
      eprint={2605.31401},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2605.31401},
}

@inproceedings{masala-etal-2024-vorbesti,
    title = "``Vorbeşti Româneşte?'' A Recipe to Train Powerful {R}omanian {LLM}s with {E}nglish Instructions",
    author = "Masala, Mihai and Ilie-Ablachim, Denis and Dima, Alexandru and Corlatescu, Dragos and Zavelca, Miruna and Olaru, Ovio and Terian, Simina and Terian, Andrei and Leordeanu, Marius and Velicu, Horia and Popescu, Marius and Dascalu, Mihai and Rebedea, Traian",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2024",
    month = nov,
    year = "2024",
    pages = "11632--11647"
}