DeepSeek-V3

llmarena.ru Created at 17.03.2025 08:06
0.677
The overall result
35
Place in the rating
Weak tasks:
83
RWSD
40
PARus
273
RCB
114
ruEthics
53
MultiQ
24
ruWorldTree
48
ruOpenBookQA
37
ruMMLU
48
ruHateSpeech
20
ruDetox
142
ruHHH
21
ruTiE
29
ruHumanEval
30
USE
35
MathLogicQA
46
ruMultiAr
60
LCS
85
ruModAr
29
ruCodeEval
+15
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.178 Accuracy
RCB 0.525 / 0.464 Accuracy F1 macro
USE 0.499 Grade norm
RWSD 0.612 Accuracy
PARus 0.93 Accuracy
ruTiE 0.896 Accuracy
MultiQ 0.581 / 0.421 F1 Exact match
CheGeKa 0.523 / 0.44 F1 Exact match
ruModAr 0.718 Exact match
MaMuRAMu 0.882 Accuracy
ruMultiAr 0.479 Exact match
ruCodeEval 0.535 / 0.751 / 0.817 Pass@k
MathLogicQA 0.857 Accuracy
ruWorldTree 0.99 / 0.99 Accuracy F1 macro
ruOpenBookQA 0.933 / 0.932 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 0.999 Accuracy
ruMMLU 0.8 Accuracy
SimpleAr 1 Exact match
ruHumanEval 0.579 / 0.786 / 0.823 Pass@k
ruHHH 0.787
ruHateSpeech 0.849
ruDetox 0.368
ruEthics
Correct God Ethical
Virtue 0.447 0.395 0.527
Law 0.431 0.373 0.496
Moral 0.478 0.412 0.539
Justice 0.402 0.347 0.464
Utilitarianism 0.368 0.318 0.441

Information about the submission

Mera version
v.1.2.0
Torch Version
2.5.1
The version of the codebase
30667dc
CUDA version
12.4
Precision of the model weights
auto
Seed
1234
Batch
1
Transformers version
4.49.0
The number of GPUs and their type
0
Architecture
local-chat-completions

Team:

llmarena.ru

Name of the ML model:

DeepSeek-V3

Model type:

Closed

API

SFT

Additional links:

Замер осуществлялся через сервис openrouter (https://openrouter.ai/deepseek/deepseek-chat)

Architecture description:

DeepSeek-V3 is a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token.

Description of the training:

We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to fully harness its capabilities.

License:

MIT License (https://huggingface.co/deepseek-ai/DeepSeek-V3/blob/main/LICENSE-CODE)

Inference parameters

Generation Parameters:
simplear - do_sample=false;until=[" \n"];do_sample=false;until=[" \n"]; \nrucodeeval - do_sample=true;temperature=0.6;until=[" \nclass"," \ndef"," \n#"," \nif"," \nprint"]; \nruhumaneval - do_sample=true;temperature=0.6;until=[" \nclass"," \ndef"," \n#"," \nif"," \nprint"];

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
DeepSeek-V3
llmarena.ru
0.767 0.733 0.833 0.333 0.433 0.467 0.4 - 0.4 0.067 0.033 0.1 0.567 0.067 0.2 0.567 0.133 0.2 0.267 0.167 0.033 0.833 0.633 0.467 0.4 0.867 0.467 0.8 0.833 0.833 0.9
Model, team Honest Helpful Harmless
DeepSeek-V3
llmarena.ru
0.803 0.763 0.793
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
DeepSeek-V3
llmarena.ru
0.815 0.542 0.895 0.885 0.869 0.891 0.835 0.83 0.895 0.735 0.702 0.667 0.7 0.824 0.894 0.795 0.84 0.903 0.8 0.84 0.685 0.871 0.7 0.78 0.741 0.91 0.592 0.75 0.78 0.64 0.81 0.884 0.81 0.889 0.845 0.88 0.74 0.935 0.748 0.773 0.904 0.912 0.793 0.899 0.933 0.806 0.931 0.719 0.649 0.807 0.8 0.911 0.872 0.933 0.9 0.848 0.922
Model, team SIM FL STA
DeepSeek-V3
llmarena.ru
0.677 0.77 0.74
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
DeepSeek-V3
llmarena.ru
0.889 0.95 0.817 0.731 0.908 0.828 0.845 0.737 0.942 0.846 0.821 0.85 0.658 0.922 0.918 0.778 0.841 0.8 0.912 0.877 0.895 0.966 0.933 0.953 0.911 0.924 0.91 0.825 0.93 0.889 0.889 0.91 0.875 0.982 0.758 0.946 0.978 0.933 0.842 0.846 0.951 0.937 0.911 1 0.948 0.911 0.948 0.932 0.938 0.912 0.911 0.957 0.797 0.792 0.698 0.871 0.933
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
DeepSeek-V3
llmarena.ru
0.447 0.431 0.478 0.402 0.368
Model, team Virtue Law Moral Justice Utilitarianism
DeepSeek-V3
llmarena.ru
0.395 0.373 0.412 0.347 0.318
Model, team Virtue Law Moral Justice Utilitarianism
DeepSeek-V3
llmarena.ru
0.527 0.496 0.539 0.464 0.441
Model, team Women Men LGBT Nationalities Migrants Other
DeepSeek-V3
llmarena.ru
0.87 0.743 0.882 0.838 0.714 0.885