--- license: cc-by-nc-4.0 language: - ro base_model: - OpenLLM-Ro/RoLlama2-7b-Base datasets: - OpenLLM-Ro/ro_sft_alpaca - OpenLLM-Ro/ro_sft_alpaca_gpt4 - OpenLLM-Ro/ro_sft_dolly - OpenLLM-Ro/ro_sft_selfinstruct_gpt4 - OpenLLM-Ro/ro_sft_norobots - OpenLLM-Ro/ro_sft_orca - OpenLLM-Ro/ro_sft_camel - OpenLLM-Ro/ro_sft_oasst - OpenLLM-Ro/ro_sft_ultrachat model-index: - name: OpenLLM-Ro/RoLlama2-7b-Instruct-2024-10-09 results: - task: type: text-generation dataset: name: RoMT-Bench type: RoMT-Bench metrics: - name: Score type: Score value: 4.43 - task: type: text-generation dataset: name: RoCulturaBench type: RoCulturaBench metrics: - name: Score type: Score value: 4.08 - task: type: text-generation dataset: name: Romanian_Academic_Benchmarks type: Romanian_Academic_Benchmarks metrics: - name: Average accuracy type: accuracy value: 44.5 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_arc_challenge type: OpenLLM-Ro/ro_arc_challenge metrics: - name: Average accuracy type: accuracy value: 44.73 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_mmlu type: OpenLLM-Ro/ro_mmlu metrics: - name: Average accuracy type: accuracy value: 40.39 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_winogrande type: OpenLLM-Ro/ro_winogrande metrics: - name: Average accuracy type: accuracy value: 63.67 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_hellaswag type: OpenLLM-Ro/ro_hellaswag metrics: - name: Average accuracy type: accuracy value: 59.12 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_gsm8k type: OpenLLM-Ro/ro_gsm8k metrics: - name: Average accuracy type: accuracy value: 13.29 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_truthfulqa type: OpenLLM-Ro/ro_truthfulqa metrics: - name: Average accuracy type: accuracy value: 45.78 - task: type: text-generation dataset: name: LaRoSeDa_binary type: LaRoSeDa_binary metrics: - name: Average macro-f1 type: macro-f1 value: 97.66 - task: type: text-generation dataset: name: LaRoSeDa_multiclass type: LaRoSeDa_multiclass metrics: - name: Average macro-f1 type: macro-f1 value: 62.41 - task: type: text-generation dataset: name: LaRoSeDa_binary_finetuned type: LaRoSeDa_binary_finetuned metrics: - name: Average macro-f1 type: macro-f1 value: 97.97 - task: type: text-generation dataset: name: LaRoSeDa_multiclass_finetuned type: LaRoSeDa_multiclass_finetuned metrics: - name: Average macro-f1 type: macro-f1 value: 60.89 - task: type: text-generation dataset: name: WMT_EN-RO type: WMT_EN-RO metrics: - name: Average bleu type: bleu value: 27.13 - task: type: text-generation dataset: name: WMT_RO-EN type: WMT_RO-EN metrics: - name: Average bleu type: bleu value: 19.39 - task: type: text-generation dataset: name: WMT_EN-RO_finetuned type: WMT_EN-RO_finetuned metrics: - name: Average bleu type: bleu value: 27.63 - task: type: text-generation dataset: name: WMT_RO-EN_finetuned type: WMT_RO-EN_finetuned metrics: - name: Average bleu type: bleu value: 39.75 - task: type: text-generation dataset: name: XQuAD type: XQuAD metrics: - name: Average exact_match type: exact_match value: 45.71 - task: type: text-generation dataset: name: XQuAD type: XQuAD metrics: - name: Average f1 type: f1 value: 65.08 - task: type: text-generation dataset: name: XQuAD_finetuned type: XQuAD_finetuned metrics: - name: Average exact_match type: exact_match value: 59.24 - task: type: text-generation dataset: name: XQuAD_finetuned type: XQuAD_finetuned metrics: - name: Average f1 type: f1 value: 74.25 - task: type: text-generation dataset: name: STS type: STS metrics: - name: Average spearman type: spearman value: 59.69 - task: type: text-generation dataset: name: STS type: STS metrics: - name: Average pearson type: pearson value: 57.16 - task: type: text-generation dataset: name: STS_finetuned type: STS_finetuned metrics: - name: Average spearman type: spearman value: 84.66 - task: type: text-generation dataset: name: STS_finetuned type: STS_finetuned metrics: - name: Average pearson type: pearson value: 85.07 - task: type: text-generation dataset: name: RoMT-Bench type: RoMT-Bench metrics: - name: First turn type: Score value: 4.92 - name: Second turn type: Score value: 3.94 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_arc_challenge type: OpenLLM-Ro/ro_arc_challenge metrics: - name: 0-shot type: accuracy value: 42.67 - name: 1-shot type: accuracy value: 44.64 - name: 3-shot type: accuracy value: 44.9 - name: 5-shot type: accuracy value: 45.16 - name: 10-shot type: accuracy value: 45.67 - name: 25-shot type: accuracy value: 45.33 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_mmlu type: OpenLLM-Ro/ro_mmlu metrics: - name: 0-shot type: accuracy value: 39.89 - name: 1-shot type: accuracy value: 40.08 - name: 3-shot type: accuracy value: 40.6 - name: 5-shot type: accuracy value: 40.99 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_winogrande type: OpenLLM-Ro/ro_winogrande metrics: - name: 0-shot type: accuracy value: 63.06 - name: 1-shot type: accuracy value: 62.98 - name: 3-shot type: accuracy value: 65.19 - name: 5-shot type: accuracy value: 63.46 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_hellaswag type: OpenLLM-Ro/ro_hellaswag metrics: - name: 0-shot type: accuracy value: 58.82 - name: 1-shot type: accuracy value: 58.44 - name: 3-shot type: accuracy value: 59.28 - name: 5-shot type: accuracy value: 59.29 - name: 10-shot type: accuracy value: 59.77 - task: type: text-generation dataset: name: OpenLLM-Ro/ro_gsm8k type: OpenLLM-Ro/ro_gsm8k metrics: - name: 0-shot type: accuracy value: 6.14 - name: 1-shot type: accuracy value: 15.01 - name: 3-shot type: accuracy value: 18.72 - task: type: text-generation dataset: name: LaRoSeDa_binary type: LaRoSeDa_binary metrics: - name: 0-shot type: macro-f1 value: 98.2 - name: 1-shot type: macro-f1 value: 96.63 - name: 3-shot type: macro-f1 value: 97.67 - name: 5-shot type: macro-f1 value: 98.13 - task: type: text-generation dataset: name: LaRoSeDa_multiclass type: LaRoSeDa_multiclass metrics: - name: 0-shot type: macro-f1 value: 63.43 - name: 1-shot type: macro-f1 value: 53.58 - name: 3-shot type: macro-f1 value: 63.78 - name: 5-shot type: macro-f1 value: 68.85 - task: type: text-generation dataset: name: WMT_EN-RO type: WMT_EN-RO metrics: - name: 0-shot type: bleu value: 20.57 - name: 1-shot type: bleu value: 29.59 - name: 3-shot type: bleu value: 29.5 - name: 5-shot type: bleu value: 28.88 - task: type: text-generation dataset: name: WMT_RO-EN type: WMT_RO-EN metrics: - name: 0-shot type: bleu value: 2.19 - name: 1-shot type: bleu value: 9.97 - name: 3-shot type: bleu value: 31.19 - name: 5-shot type: bleu value: 34.23 - task: type: text-generation dataset: name: XQuAD_EM type: XQuAD_EM metrics: - name: 0-shot type: exact_match value: 40.25 - name: 1-shot type: exact_match value: 46.47 - name: 3-shot type: exact_match value: 47.56 - name: 5-shot type: exact_match value: 48.57 - task: type: text-generation dataset: name: XQuAD_F1 type: XQuAD_F1 metrics: - name: 0-shot type: f1 value: 62.24 - name: 1-shot type: f1 value: 65.33 - name: 3-shot type: f1 value: 65.89 - name: 5-shot type: f1 value: 66.86 - task: type: text-generation dataset: name: STS type: STS metrics: - name: 0-shot type: spearman value: 55.44 - name: 1-shot type: spearman value: 61.98 - name: 3-shot type: spearman value: 61.65 - task: type: text-generation dataset: name: STS type: STS metrics: - name: 0-shot type: pearson value: 56.18 - name: 1-shot type: pearson value: 58.37 - name: 3-shot type: pearson value: 56.94 --- # Model Card for Model ID RoLlama2 is a family of pretrained and fine-tuned generative text models for Romanian. This is the repository for the **instruct 7B model**. Links to other models can be found at the bottom of this page. ## Model Details ### Model Description OpenLLM 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:** [RoLlama2-7b-Base](https://huggingface.co/OpenLLM-Ro/RoLlama2-7b-Base) - **Trained using:** [RoAlpaca](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_alpaca), [RoAlpacaGPT4](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_alpaca_gpt4), [RoDolly](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_dolly), [RoSelfInstruct](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_selfinstruct_gpt4), [RoNoRobots](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_norobots), [RoOrca](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_orca), [RoCamel](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_camel), [RoOpenAssistant](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_oasst), [RoUltraChat](https://huggingface.co/datasets/OpenLLM-Ro/ro_sft_ultrachat) ### Model Sources - **Repository:** https://github.com/OpenLLM-Ro/LLaMA-Factory - **Paper:** https://arxiv.org/abs/2406.18266 ## Intended Use ### Intended Use Cases RoLlama2 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. ```python from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("OpenLLM-Ro/RoLlama2-7b-Instruct") model = AutoModelForCausalLM.from_pretrained("OpenLLM-Ro/RoLlama2-7b-Instruct") instruction = "Care este cel mai înalt vârf muntos din România?" 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) 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-2-7b-chat
36.84
37.03
33.80
55.87
45.36
4.90
44.09
RoLlama2-7b-Instruct-2024-05-14
45.71
43.66
39.70
70.34
57.36
18.78
44.44
RoLlama2-7b-Instruct-2024-10-09
44.50
44.73
40.39
63.67
59.12
13.29
45.78
## 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-2-7b-chat
87.78
52.81
97.27
82.02
15.55
28.53
19.99
31.48
RoLlama2-7b-Instruct-2024-05-14
97.48
65.26
98.83
87.28
27.38
10.32
27.59
40.13
RoLlama2-7b-Instruct-2024-10-09
97.66
62.41
97.97
60.89
27.13
19.39
27.63
39.75
XQuAD
STS
Few-shot
Finetuned
Few-shot
Finetuned
Model
(EM)
(F1)
(EM)
(F1)
(Spearman)
(Pearson)
(Spearman)
(Pearson)
Llama-2-7b-chat
32.35
54.00
60.34
75.98
32.56
31.99
74.08
72.64
RoLlama2-7b-Instruct-2024-05-14
44.52
64.75
54.96
70.20
65.50
67.79
84.44
84.76
RoLlama2-7b-Instruct-2024-10-09
45.71
65.08
59.24
74.25
59.69
57.16
84.66
85.07
## Romanian MT-Bench
Model
Average
1st turn
2nd turn
Answers in Ro
Llama-2-7b-chat
1.08
1.44
0.73
45/160
RoLlama2-7b-Instruct-2024-05-14
3.86
4.67
3.04
160/160
RoLlama2-7b-Instruct-2024-10-09
4.43
4.92
3.94
160/160
## RoCulturaBench
Model
Average
Answers in Ro
Llama-2-7b-chat
1.21
33/100
RoLlama2-7b-Instruct-2024-05-14
3.77
100/100
RoLlama2-7b-Instruct-2024-10-09
4.08
100/100
## RoLlama2 Model Family | Model | Link | |--------------------|:--------:| |RoLlama2-7b-Base | [link](https://huggingface.co/OpenLLM-Ro/RoLlama2-7b-Base) | |*RoLlama2-7b-Instruct*| [link](https://huggingface.co/OpenLLM-Ro/RoLlama2-7b-Instruct) | ## Citation ``` @misc{masala2024vorbecstiromanecsterecipetrain, title={"Vorbe\c{s}ti Rom\^ane\c{s}te?" A Recipe to Train Powerful Romanian LLMs with English Instructions}, author={Mihai Masala and Denis C. Ilie-Ablachim and Alexandru Dima and Dragos Corlatescu and Miruna Zavelca and Ovio Olaru and Simina Terian-Dan and Andrei Terian-Dan and Marius Leordeanu and Horia Velicu and Marius Popescu and Mihai Dascalu and Traian Rebedea}, year={2024}, eprint={2406.18266}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2406.18266}, } ```