RoLlama3.1-8b-Instruct-DPO/README.md

license, language, base_model, datasets, model-index

license	language	base_model	datasets	model-index
cc-by-nc-4.0	ro	OpenLLM-Ro/RoLlama3.1-8b-Instruct-2025-04-23	OpenLLM-Ro/ro_dpo_helpsteer OpenLLM-Ro/ro_dpo_ultrafeedback OpenLLM-Ro/ro_dpo_magpie OpenLLM-Ro/ro_dpo_argilla_magpie OpenLLM-Ro/ro_dpo_helpsteer2	name results OpenLLM-Ro/RoLlama3.1-8b-Instruct-DPO-2025-04-23 task dataset metrics type text-generation name type RoMT-Bench RoMT-Bench name type value Score Score 7.00 task dataset metrics type text-generation name type RoCulturaBench RoCulturaBench name type value Score Score 4.73 task dataset metrics type text-generation name type Romanian_Academic_Benchmarks Romanian_Academic_Benchmarks name type value Average accuracy accuracy 53.76 task dataset metrics type text-generation name type OpenLLM-Ro/ro_arc_challenge OpenLLM-Ro/ro_arc_challenge name type value Average accuracy accuracy 51.09 task dataset metrics type text-generation name type OpenLLM-Ro/ro_mmlu OpenLLM-Ro/ro_mmlu name type value Average accuracy accuracy 56.22 task dataset metrics type text-generation name type OpenLLM-Ro/ro_winogrande OpenLLM-Ro/ro_winogrande name type value Average accuracy accuracy 66.77 task dataset metrics type text-generation name type OpenLLM-Ro/ro_hellaswag OpenLLM-Ro/ro_hellaswag name type value Average accuracy accuracy 59.38 task dataset metrics type text-generation name type OpenLLM-Ro/ro_gsm8k OpenLLM-Ro/ro_gsm8k name type value Average accuracy accuracy 31.54 task dataset metrics type text-generation name type OpenLLM-Ro/ro_truthfulqa OpenLLM-Ro/ro_truthfulqa name type value Average accuracy accuracy 57.56 task dataset metrics type text-generation name type LaRoSeDa_binary LaRoSeDa_binary name type value Average macro-f1 macro-f1 96.87 task dataset metrics type text-generation name type LaRoSeDa_multiclass LaRoSeDa_multiclass name type value Average macro-f1 macro-f1 60.75 task dataset metrics type text-generation name type WMT_EN-RO WMT_EN-RO name type value Average bleu bleu 20.30 task dataset metrics type text-generation name type WMT_RO-EN WMT_RO-EN name type value Average bleu bleu 18.57 task dataset metrics type text-generation name type XQuAD XQuAD name type value Average exact_match exact_match 9.22 task dataset metrics type text-generation name type XQuAD XQuAD name type value Average f1 f1 22.75 task dataset metrics type text-generation name type STS STS name type value Average spearman spearman 30.82 task dataset metrics type text-generation name type STS STS name type value Average pearson pearson 20.25 task dataset metrics type text-generation name type RoMT-Bench RoMT-Bench name type value First turn Score 7.30 name type value Second turn Score 6.70 task dataset metrics type text-generation name type OpenLLM-Ro/ro_arc_challenge OpenLLM-Ro/ro_arc_challenge name type value 0-shot accuracy 51.59 name type value 1-shot accuracy 52.10 name type value 3-shot accuracy 50.99 name type value 5-shot accuracy 50.81 name type value 10-shot accuracy 49.70 name type value 25-shot accuracy 51.33 task dataset metrics type text-generation name type OpenLLM-Ro/ro_mmlu OpenLLM-Ro/ro_mmlu name type value 0-shot accuracy 56.88 name type value 1-shot accuracy 55.61 name type value 3-shot accuracy 56.06 name type value 5-shot accuracy 56.31 task dataset metrics type text-generation name type OpenLLM-Ro/ro_winogrande OpenLLM-Ro/ro_winogrande name type value 0-shot accuracy 65.67 name type value 1-shot accuracy 66.30 name type value 3-shot accuracy 67.40 name type value 5-shot accuracy 67.72 task dataset metrics type text-generation name type OpenLLM-Ro/ro_hellaswag OpenLLM-Ro/ro_hellaswag name type value 0-shot accuracy 60.53 name type value 1-shot accuracy 60.37 name type value 3-shot accuracy 58.20 name type value 5-shot accuracy 58.18 name type value 10-shot accuracy 59.61 task dataset metrics type text-generation name type OpenLLM-Ro/ro_gsm8k OpenLLM-Ro/ro_gsm8k name type value 1-shot accuracy 25.09 name type value 3-shot accuracy 30.02 name type value 5-shot accuracy 39.50 task dataset metrics type text-generation name type LaRoSeDa_binary LaRoSeDa_binary name type value 0-shot macro-f1 95.39 name type value 1-shot macro-f1 95.90 name type value 3-shot macro-f1 98.00 name type value 5-shot macro-f1 98.17 task dataset metrics type text-generation name type LaRoSeDa_multiclass LaRoSeDa_multiclass name type value 0-shot macro-f1 60.30 name type value 1-shot macro-f1 64.73 name type value 3-shot macro-f1 58.69 name type value 5-shot macro-f1 59.30 task dataset metrics type text-generation name type WMT_EN-RO WMT_EN-RO name type value 0-shot bleu 5.46 name type value 1-shot bleu 26.08 name type value 3-shot bleu 25.90 name type value 5-shot bleu 23.76 task dataset metrics type text-generation name type WMT_RO-EN WMT_RO-EN name type value 0-shot bleu 2.74 name type value 1-shot bleu 20.95 name type value 3-shot bleu 31.53 name type value 5-shot bleu 19.05 task dataset metrics type text-generation name type XQuAD_EM XQuAD_EM name type value 0-shot exact_match 12.27 name type value 1-shot exact_match 17.98 name type value 3-shot exact_match 5.04 name type value 5-shot exact_match 1.60 task dataset metrics type text-generation name type XQuAD_F1 XQuAD_F1 name type value 0-shot f1 26.24 name type value 1-shot f1 32.54 name type value 3-shot f1 18.00 name type value 5-shot f1 14.22 task dataset metrics type text-generation name type STS_Spearman STS_Spearman name type value 1-shot spearman 76.70 name type value 3-shot spearman 2.82 name type value 5-shot spearman 12.95 task dataset metrics type text-generation name type STS_Pearson STS_Pearson name type value 1-shot pearson 77.30 name type value 3-shot pearson -14.56 name type value 5-shot pearson -1.99

Model Card for Model ID

Built with Meta Llama 3.1

This model points/is identical to RoLlama3.1-8b-Instruct-DPO-2025-04-23.

RoLlama3.1 is a family of pretrained and fine-tuned generative text models for Romanian. This is the repository for the human aligned instruct 8B model. Links to other models can be found at the bottom of this page.

Model Details

Model Description

OpenLLM-Ro represents the first open-source effort to build a LLM specialized for Romanian. OpenLLM-Ro developed and publicly releases a collection of Romanian LLMs, both in the form of foundational model and instruct and chat variants.

• Developed by: OpenLLM-Ro

• Language(s): Romanian
• License: cc-by-nc-4.0
• Finetuned from model: RoLlama3.1-8b-Instruct-2025-04-23
• Trained using: RoHelpSteer, RoUltraFeedback, RoMagpieDPO, RoArgillaMagpie, RoHelpSteer2

Model Sources

• Repository: https://github.com/OpenLLM-Ro/LLaMA-Factory
• Paper: https://arxiv.org/abs/2406.18266

Intended Use

Intended Use Cases

RoLlama3.1 is intented for research use in Romanian. Base models can be adapted for a variety of natural language tasks while instruction and chat tuned models are intended for assistant-like chat.

Out-of-Scope Use

Use in any manner that violates the license, any applicable laws or regluations, use in languages other than Romanian.

How to Get Started with the Model

Use the code below to get started with the model.

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("OpenLLM-Ro/RoLlama3.1-8b-Instruct-DPO")
model = AutoModelForCausalLM.from_pretrained("OpenLLM-Ro/RoLlama3.1-8b-Instruct-DPO")

instruction = "Ce jocuri de societate pot juca cu prietenii mei?"
chat = [
        {"role": "system", "content": "Ești un asistent folositor, respectuos și onest. Încearcă să ajuți cât mai mult prin informațiile oferite, excluzând răspunsuri toxice, rasiste, sexiste, periculoase și ilegale."},
        {"role": "user", "content": instruction},
        ]
prompt = tokenizer.apply_chat_template(chat, tokenize=False, system_message="")

inputs = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
outputs = model.generate(input_ids=inputs, max_new_tokens=128)
print(tokenizer.decode(outputs[0]))

Academic Benchmarks

Model	Average	ARC	MMLU	Winogrande	Hellaswag	GSM8k	TruthfulQA
Llama-3.1-8B-Instruct	49.87	42.86	53.73	59.71	56.82	35.56	50.54
RoLlama3.1-8b-Instruct-2024-10-09	53.03	47.69	54.57	65.84	59.94	44.30	45.82
RoLlama3.1-8b-Instruct-2025-04-23	53.36	48.97	55.17	66.52	60.73	42.03	46.71
RoLlama3.1-8b-Instruct-DPO-2024-10-09	52.74	44.84	55.06	65.87	58.67	44.17	47.82
RoLlama3.1-8b-Instruct-DPO-2025-04-23	53.76	51.09	56.22	66.77	59.38	31.54	57.56

Downstream tasks

	LaRoSeDa				WMT
	Few-shot		Finetuned		Few-shot		Finetuned
Model	Binary (Macro F1)	Multiclass (Macro F1)	Binary (Macro F1)	Multiclass (Macro F1)	EN-RO (Bleu)	RO-EN (Bleu)	EN-RO (Bleu)	RO-EN (Bleu)
Llama-3.1-8B-Instruct	95.74	59.49	98.57	82.41	19.01	27.77	29.02	39.80
RoLlama3.1-8b-Instruct-2024-10-09	94.56	60.10	95.12	87.53	21.88	23.99	28.27	40.44
RoLlama3.1-8b-Instruct-2025-04-23	95.32	60.84	-	-	23.18	25.11	-	-
RoLlama3.1-8b-Instruct-DPO-2024-10-09	96.10	55.37	-	-	21.29	21.86	-	-
RoLlama3.1-8b-Instruct-DPO-2025-04-23	96.87	60.75	-	-	20.30	18.57	-	-

	XQuAD				STS
	Few-shot		Finetuned		Few-shot		Finetuned
Model	(EM)	(F1)	(EM)	(F1)	(Spearman)	(Pearson)	(Spearman)	(Pearson)
Llama-3.1-8B-Instruct	44.96	64.45	69.50	84.31	72.11	71.64	84.59	84.96
RoLlama3.1-8b-Instruct-2024-10-09	13.59	23.56	49.41	62.93	75.89	76.00	86.86	87.05
RoLlama3.1-8b-Instruct-2025-04-23	10.74	19.75	-	-	73.53	74.93	-	-
RoLlama3.1-8b-Instruct-DPO-2024-10-09	21.58	36.54	-	-	78.01	77.98	-	-
RoLlama3.1-8b-Instruct-DPO-2025-04-23	9.22	22.75	-	-	30.82	20.25	-	-

MT-Bench

Model	Average	1st turn	2nd turn	Answers in Ro
Llama-3.1-8B-Instruct	5.69	5.85	5.53	160/160
RoLlama3.1-8b-Instruct-2024-10-09	5.42	5.95	4.89	160/160
RoLlama3.1-8b-Instruct-2025-04-23	6.43	6.78	6.09	160/160
RoLlama3.1-8b-Instruct-DPO-2024-10-09	6.21	6.74	5.69	160/160
RoLlama3.1-8b-Instruct-DPO-2025-04-23	7.00	7.30	6.70	160/160

RoCulturaBench

Model	Average	Answers in Ro
Llama-3.1-8B-Instruct	3.54	100/100
RoLlama3.1-8b-Instruct-2024-10-09	3.55	100/100
RoLlama3.1-8b-Instruct-2025-04-23	4.28	100/100
RoLlama3.1-8b-Instruct-DPO-2024-10-09	4.42	100/100
RoLlama3.1-8b-Instruct-DPO-2025-04-23	4.73	100/100

RoLlama3.1 Model Family

Model	Link
RoLlama3.1-8b-Instruct-2024-10-09	link
RoLlama3.1-8b-Instruct-2025-04-23	link
RoLlama3.1-8b-Instruct-DPO-2024-10-09	link
RoLlama3.1-8b-Instruct-DPO-2025-04-23	link

Citation

@inproceedings{masala-etal-2024-vorbesti,
    title = "``Vorbe\c{s}ti Rom{\^a}ne\c{s}te?'' 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 Georgian and Zavelca, Miruna-Andreea and Olaru, Ovio and Terian, Simina-Maria and Terian, Andrei and Leordeanu, Marius and Velicu, Horia and Popescu, Marius and Dascalu, Mihai and Rebedea, Traian",
    editor = "Al-Onaizan, Yaser and Bansal, Mohit and Chen, Yun-Nung",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2024",
    month = nov,
    year = "2024",
    address = "Miami, Florida, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.findings-emnlp.681/",
    doi = "10.18653/v1/2024.findings-emnlp.681",
    pages = "11632--11647"
}