Qwen2.5-72B-Instruct

MERA Created at 04.11.2024 17:50
0.623
The overall result
51
Place in the rating
Weak tasks:
33
RWSD
128
RCB
32
ruEthics
106
MultiQ
34
ruWorldTree
27
ruOpenBookQA
116
CheGeKa
51
ruMMLU
43
ruHateSpeech
33
ruHHH
32
ruTiE
66
ruHumanEval
106
USE
68
MathLogicQA
68
ruMultiAr
60
SimpleAr
114
LCS
42
BPS
44
ruModAr
43
MaMuRAMu
100
ruCodeEval
+17
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.148 Accuracy
RCB 0.562 / 0.526 Accuracy F1 macro
USE 0.294 Grade norm
RWSD 0.696 Accuracy
PARus 0.95 Accuracy
ruTiE 0.878 Accuracy
MultiQ 0.509 / 0.363 F1 Exact match
CheGeKa 0.313 / 0.243 F1 Exact match
ruModAr 0.854 Exact match
MaMuRAMu 0.847 Accuracy
ruMultiAr 0.416 Exact match
ruCodeEval 0.21 / 0.427 / 0.494 Pass@k
MathLogicQA 0.69 Accuracy
ruWorldTree 0.989 / 0.989 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 0.994 Accuracy
ruMMLU 0.774 Accuracy
SimpleAr 0.995 Exact match
ruHumanEval 0.312 / 0.545 / 0.622 Pass@k
ruHHH 0.871
ruHateSpeech 0.853
ruDetox 0.371
ruEthics
Correct God Ethical
Virtue 0.47 0.429 0.515
Law 0.462 0.416 0.485
Moral 0.504 0.459 0.548
Justice 0.407 0.378 0.466
Utilitarianism 0.397 0.375 0.428

Information about the submission

Mera version
v.1.2.0
Torch Version
2.4.0
The version of the codebase
fea61e4
CUDA version
12.1
Precision of the model weights
bfloat16bfloat16
Seed
1234
Batch
1
Transformers version
4.45.1
The number of GPUs and their type
8 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Qwen2.5-72B-Instruct

Model size

72.7B

Model type:

Opened

SFT

Additional links:

https://qwenlm.github.io/blog/qwen2.5-llm/

Architecture description:

Type: Causal Language Models Training Stage: Pretraining & Post-training Architecture: transformers with RoPE, SwiGLU, RMSNorm, and Attention QKV bias Number of Parameters: 72.7B Number of Paramaters (Non-Embedding): 70.0B Number of Layers: 80 Number of Attention Heads (GQA): 64 for Q and 8 for KV Context Length: Full 131,072 tokens and generation 8192 tokens

Description of the training:

-

Pretrain data:

The size of the pre-training dataset is expanded from 7 trillion tokens to a maximum of 18 trillion tokens.

License:

qwen

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, rummlu, ruworldtree, ruopenbookqa, rumultiar, use, rwsd, mamuramu, multiq, rumodar, ruethics, ruhhh, ruhumaneval, rucodeeval - 32768 \nrutie - 20000

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.

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
Qwen2.5-72B-Instruct
MERA
0.333 0.667 0.5 0.167 0.1 0.433 0.167 - 0.1 0.067 0.033 0 0.333 0.1 0.1 0.333 0.1 0 0 0 0.1 0.633 0.333 0.3 0.067 0.625 0.367 0.233 0.533 0.467 0.6
Model, team Honest Helpful Harmless
Qwen2.5-72B-Instruct
MERA
0.852 0.898 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
Qwen2.5-72B-Instruct
MERA
0.733 0.566 0.888 0.902 0.859 0.861 0.816 0.788 0.883 0.749 0.64 0.627 0.61 0.769 0.879 0.795 0.81 0.931 0.667 0.84 0.621 0.883 0.65 0.757 0.714 0.91 0.576 0.694 0.771 0.62 0.81 0.851 0.773 0.919 0.823 0.855 0.64 0.923 0.695 0.744 0.894 0.86 0.71 0.836 0.912 0.75 0.926 0.644 0.589 0.789 0.77 0.886 0.859 0.912 0.93 0.867 0.896
Model, team SIM FL STA
Qwen2.5-72B-Instruct
MERA
0.731 0.704 0.743
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
Qwen2.5-72B-Instruct
MERA
0.756 0.911 0.767 0.75 0.868 0.862 0.759 0.772 0.846 0.815 0.833 0.8 0.658 0.884 0.848 0.802 0.785 0.778 0.825 0.842 0.895 0.932 0.867 0.858 0.867 0.894 0.872 0.719 0.947 0.867 0.844 0.897 0.875 0.912 0.773 0.875 0.911 0.844 0.754 0.815 0.902 0.873 0.822 1 0.879 0.911 0.948 0.955 0.892 0.947 0.867 0.928 0.848 0.792 0.581 0.854 0.922
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-72B-Instruct
MERA
0.47 0.462 0.504 0.407 0.397
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-72B-Instruct
MERA
0.429 0.416 0.459 0.378 0.375
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-72B-Instruct
MERA
0.515 0.485 0.548 0.466 0.428
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2.5-72B-Instruct
MERA
0.852 0.743 0.824 0.892 0.571 0.934