Qwen2-72B-Instruct

MERA Created at 22.09.2024 21:58
0.57
The overall result
90
Place in the rating
In the top by tasks:
4
ruEthics
Weak tasks:
54
RWSD
313
RCB
54
MultiQ
51
ruWorldTree
29
ruOpenBookQA
104
CheGeKa
68
ruMMLU
31
ruHateSpeech
46
ruDetox
60
ruHHH
43
ruTiE
298
ruHumanEval
77
USE
88
MathLogicQA
84
ruMultiAr
69
SimpleAr
78
LCS
66
BPS
140
ruModAr
57
MaMuRAMu
+16
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.164 Accuracy
RCB 0.511 / 0.484 Accuracy F1 macro
USE 0.325 Grade norm
RWSD 0.658 Accuracy
PARus 0.944 Accuracy
ruTiE 0.87 Accuracy
MultiQ 0.58 / 0.447 F1 Exact match
CheGeKa 0.324 / 0.267 F1 Exact match
ruModAr 0.6 Exact match
MaMuRAMu 0.838 Accuracy
ruMultiAr 0.387 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.626 Accuracy
ruWorldTree 0.985 / 0.789 Accuracy F1 macro
ruOpenBookQA 0.945 / 0.945 Accuracy F1 macro

Evaluation on open tasks:

Go to the ratings by subcategory

The table will scroll to the left

Task name Result Metric
BPS 0.988 Accuracy
ruMMLU 0.756 Accuracy
SimpleAr 0.994 Exact match
ruHumanEval 0.006 / 0.006 / 0.006 Pass@k
ruHHH 0.86
ruHateSpeech 0.868
ruDetox 0.335
ruEthics
Correct God Ethical
Virtue 0.55 0.532 0.702
Law 0.582 0.533 0.707
Moral 0.621 0.569 0.768
Justice 0.499 0.474 0.645
Utilitarianism 0.48 0.461 0.603

Information about the submission

Mera version
v.1.2.0
Torch Version
2.4.0
The version of the codebase
9b26db97
CUDA version
12.1
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.44.2
The number of GPUs and their type
8 x NVIDIA H100 80GB HBM3
Architecture
vllm

Team:

MERA

Name of the ML model:

Qwen2-72B-Instruct

Model size

72.7B

Model type:

Opened

SFT

Additional links:

https://qwenlm.github.io/blog/qwen2/ https://arxiv.org/html/2407.10671v2

Architecture description:

Qwen 2 is a decoder-only language model.

Description of the training:

The model was pretrained with a large amount of data and post-trained the models with both supervised finetuning and direct preference optimization.

Pretrain data:

-

License:

tongyi-qianwen

Inference parameters

Generation Parameters:
simplear - do_sample=false;until=["\n"]; \nchegeka - do_sample=false;until=["\n"]; \nrudetox - do_sample=false;until=["\n"]; \nrumultiar - do_sample=false;until=["\n"]; \nuse - do_sample=false;until=["\n","."]; \nmultiq - do_sample=false;until=["\n"]; \nrumodar - do_sample=false;until=["\n"]; \nruhumaneval - do_sample=true;until=["\nclass","\ndef","\n#","\nif","\nprint"];temperature=0.6; \nrucodeeval - do_sample=true;until=["\nclass","\ndef","\n#","\nif","\nprint"];temperature=0.6;

The size of the context:
simplear, bps, lcs, chegeka, mathlogicqa, parus, rcb, rudetox, ruhatespeech, ruworldtree, ruopenbookqa, rumultiar, use, rwsd, mamuramu, multiq, rumodar, ruethics, ruhumaneval, rucodeeval, rummlu, ruhhh - 32768 \nrutie - 2000

System prompt:
Реши задачу по инструкции ниже. Не давай никаких объяснений и пояснений к своему ответу. Не пиши ничего лишнего. Пиши только то, что указано в инструкции. Если по инструкции нужно решить пример, то напиши только числовой ответ без хода решения и пояснений. Если по инструкции нужно вывести букву, цифру или слово, выведи только его. Если по инструкции нужно выбрать один из вариантов ответа и вывести букву или цифру, которая ему соответствует, то выведи только эту букву или цифру, не давай никаких пояснений, не добавляй знаки препинания, только 1 символ в ответе. Если по инструкции нужно дописать код функции на языке Python, пиши сразу код, соблюдая отступы так, будто ты продолжаешь функцию из инструкции, не давай пояснений, не пиши комментарии, используй только аргументы из сигнатуры функции в инструкции, не пробуй считывать данные через функцию input. Не извиняйся, не строй диалог. Выдавай только ответ и ничего больше.

Description of the template:
{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system \nYou are a helpful assistant.<|im_end|> \n' }}{% endif %}{{'<|im_start|>' + message['role'] + ' \n' + message['content'] + '<|im_end|>' + ' \n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant \n' }}{% endif %}

Ratings by subcategory

Metric: Grade Norm
Model, team 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 8_0 8_1 8_2 8_3 8_4
Qwen2-72B-Instruct
MERA
0.533 0.633 0.8 0.1 0.067 0.533 0.2 - 0.167 0.067 0.033 0.033 0.233 0.1 0.133 0.417 0 0 0.067 0.067 0.133 0.767 0.433 0.2 0.233 0.708 0.2 0.233 0.533 0.367 0.533
Model, team Honest Helpful Harmless
Qwen2-72B-Instruct
MERA
0.852 0.864 0.862
Model, team Anatomy Virology Astronomy Marketing Nutrition Sociology Management Philosophy Prehistory Human aging Econometrics Formal logic Global facts Jurisprudence Miscellaneous Moral disputes Business ethics Biology (college) Physics (college) Human Sexuality Moral scenarios World religions Abstract algebra Medicine (college) Machine learning Medical genetics Professional law PR Security studies Chemistry (школьная) Computer security International law Logical fallacies Politics Clinical knowledge Conceptual_physics Math (college) Biology (high school) Physics (high school) Chemistry (high school) Geography (high school) Professional medicine Electrical engineering Elementary mathematics Psychology (high school) Statistics (high school) History (high school) Math (high school) Professional accounting Professional psychology Computer science (college) World history (high school) Macroeconomics Microeconomics Computer science (high school) European history Government and politics
Qwen2-72B-Instruct
MERA
0.763 0.554 0.901 0.893 0.84 0.866 0.796 0.791 0.855 0.762 0.649 0.579 0.66 0.833 0.89 0.772 0.79 0.882 0.633 0.802 0.542 0.86 0.59 0.711 0.714 0.87 0.575 0.713 0.776 0.56 0.77 0.851 0.748 0.889 0.789 0.85 0.52 0.91 0.662 0.719 0.889 0.824 0.71 0.735 0.91 0.718 0.902 0.604 0.603 0.774 0.77 0.873 0.828 0.903 0.92 0.873 0.907
Model, team SIM FL STA
Qwen2-72B-Instruct
MERA
0.629 0.733 0.756
Model, team Anatomy Virology Astronomy Marketing Nutrition Sociology Managment Philosophy Pre-History Gerontology Econometrics Formal logic Global facts Jurisprudence Miscellaneous Moral disputes Business ethics Bilology (college) Physics (college) Human sexuality Moral scenarios World religions Abstract algebra Medicine (college) Machine Learning Genetics Professional law PR Security Chemistry (college) Computer security International law Logical fallacies Politics Clinical knowledge Conceptual physics Math (college) Biology (high school) Physics (high school) Chemistry (high school) Geography (high school) Professional medicine Electrical Engineering Elementary mathematics Psychology (high school) Statistics (high school) History (high school) Math (high school) Professional Accounting Professional psychology Computer science (college) World history (high school) Macroeconomics Microeconomics Computer science (high school) Europe History Government and politics
Qwen2-72B-Instruct
MERA
0.622 0.921 0.8 0.713 0.895 0.897 0.793 0.737 0.923 0.831 0.808 0.792 0.642 0.814 0.807 0.765 0.841 0.844 0.772 0.825 0.912 0.949 0.911 0.917 0.8 0.924 0.872 0.772 0.93 0.822 0.822 0.91 0.857 0.93 0.758 0.875 0.844 0.844 0.772 0.831 0.91 0.905 0.867 0.933 0.914 0.867 0.914 0.886 0.862 0.947 0.822 0.913 0.785 0.701 0.651 0.754 0.911
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-72B-Instruct
MERA
0.55 0.582 0.621 0.499 0.48
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-72B-Instruct
MERA
0.532 0.533 0.569 0.474 0.461
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-72B-Instruct
MERA
0.702 0.707 0.768 0.645 0.603
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2-72B-Instruct
MERA
0.889 0.743 0.824 0.865 0.857 0.918