Safetensors
Romanian
mistral
Eval Results

Model Card for Model ID

This model points/is identical to RoMistral-7b-Instruct-DPO-2024-10-09.

RoMistral is a family of pretrained and fine-tuned generative text models for Romanian. This is the repository for the human aligned instruct 7B 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.

Model Sources

Intended Use

Intended Use Cases

RoMistral 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/RoMistral-7b-Instruct-DPO")
model = AutoModelForCausalLM.from_pretrained("OpenLLM-Ro/RoMistral-7b-Instruct-DPO")

instruction = "Ce jocuri de societate pot juca cu prietenii mei?"
chat = [
        {"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
Mistral-7B-Instruct-v0.2
47.40
46.29
47.00
58.78
54.27
13.47
64.59
RoMistral-7b-Instruct-2024-05-17
52.54
50.41
51.61
66.48
60.27
34.19
52.30
RoMistral-7b-Instruct-2024-10-09
52.91
52.27
49.33
70.03
62.88
32.42
50.51
RoMistral-7b-Instruct-DPO-2024-10-09
51.95
50.73
47.88
68.41
62.27
32.27
50.12

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)
Mistral-7B-Instruct-v0.2
96.97
56.66
98.83
87.32
18.60
33.99
26.19
39.88
RoMistral-7b-Instruct-2024-05-17
97.36
67.55
98.80
88.28
27.93
13.21
28.72
40.86
RoMistral-7b-Instruct-2024-10-09
95.56
67.83
99.00
87.57
28.28
6.10
27.70
40.36
RoMistral-7b-Instruct-DPO-2024-10-09
82.13
65.24
-
-
26.25
6.09
-
-
XQuAD
STS
Few-shot
Finetuned
Few-shot
Finetuned
Model
(EM)
(F1)
(EM)
(F1)
(Spearman)
(Pearson)
(Spearman)
(Pearson)
Mistral-7B-Instruct-v0.2
27.92
50.71
65.46
79.73
62.62
60.86
84.92
85.44
RoMistral-7b-Instruct-2024-05-17
43.66
63.70
55.04
72.31
77.43
78.43
87.25
87.79
RoMistral-7b-Instruct-2024-10-09
41.09
63.21
47.56
62.69
78.47
77.24
87.28
87.88
RoMistral-7b-Instruct-DPO-2024-10-09
23.40
45.80
-
-
77.33
76.60
-
-

MT-Bench

Model
Average
1st turn
2nd turn
Answers in Ro
Mistral-7B-Instruct-v0.2
5.03
5.05
5.00
154/160
RoMistral-7b-Instruct-2024-05-17
4.99
5.46
4.53
160/160
RoMistral-7b-Instruct-2024-10-09
5.29
5.86
4.72
160/160
RoMistral-7b-Instruct-DPO-2024-10-09
5.88
6.44
5.33
160/160

RoCulturaBench

Model
Average
Answers in Ro
Mistral-7B-Instruct-v0.2
3.68
97/100
RoMistral-7b-Instruct-2024-05-17
3.38
100/100
RoMistral-7b-Instruct-2024-10-09
3.99
100/100
RoMistral-7b-Instruct-DPO-2024-10-09
4.72
100/100

RoMistral Model Family

Model Link
RoMistral-7b-Instruct-2024-05-17 link
RoMistral-7b-Instruct-2024-10-09 link
RoMistral-7b-Instruct-DPO-2024-10-09 link

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}, 
}
Downloads last month
130
Safetensors
Model size
7.24B params
Tensor type
BF16
·
Inference API
Unable to determine this model's library. Check the docs .

Model tree for OpenLLM-Ro/RoMistral-7b-Instruct-DPO

Finetuned
(2)
this model

Dataset used to train OpenLLM-Ro/RoMistral-7b-Instruct-DPO

Collection including OpenLLM-Ro/RoMistral-7b-Instruct-DPO

Evaluation results