DeepSeek-R1-Distill-Qwen-32B

НГУ Created at 06.06.2025 09:02
0.618
The overall result
53
Place in the rating
In the top by tasks:
2
PARus
The task is one of the main ones
9
ruEthics
Weak tasks:
26
RWSD
106
RCB
113
MultiQ
173
ruWorldTree
103
ruOpenBookQA
144
CheGeKa
93
ruMMLU
24
ruHateSpeech
77
ruDetox
46
ruHHH
90
ruTiE
250
ruHumanEval
83
USE
32
MathLogicQA
30
ruMultiAr
417
SimpleAr
58
LCS
89
BPS
54
ruModAr
107
MaMuRAMu
+16
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.182 Accuracy
RCB 0.566 / 0.523 Accuracy F1 macro
USE 0.319 Grade norm
RWSD 0.742 Accuracy
PARus 0.962 Accuracy
ruTiE 0.822 Accuracy
MultiQ 0.503 / 0.331 F1 Exact match
CheGeKa 0.208 / 0.163 F1 Exact match
ruModAr 0.831 Exact match
MaMuRAMu 0.791 Accuracy
ruMultiAr 0.909 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.904 Accuracy
ruWorldTree 0.937 / 0.765 Accuracy F1 macro
ruOpenBookQA 0.898 / 0.725 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 0.983 Accuracy
ruMMLU 0.718 Accuracy
SimpleAr 0.817 Exact match
ruHumanEval 0.012 / 0.012 / 0.012 Pass@k
ruHHH 0.865
ruHateSpeech 0.875
ruDetox 0.313
ruEthics
Correct God Ethical
Virtue 0.536 0.544 0.554
Law 0.546 0.522 0.539
Moral 0.576 0.569 0.593
Justice 0.483 0.477 0.529
Utilitarianism 0.427 0.45 0.454

Information about the submission

Mera version
v.1.2.0
Torch Version
2.6.0
The version of the codebase
9e50515
CUDA version
12.4
Precision of the model weights
bf16
Seed
1234
Batch
1
Transformers version
4.51.3
The number of GPUs and their type
8 x NVIDIA A800 80GB PCIe
Architecture
local-chat-completions

Team:

НГУ

Name of the ML model:

DeepSeek-R1-Distill-Qwen-32B

Model size

32.0B

Model type:

Opened

SFT

Architecture description:

Causal Language Model; Training Stage: Pretraining, fine-tuned based on open-source models, using samples generated by DeepSeek-R1; Architecture: transformers with RoPE, SwiGLU, RMSNorm, and Attention QKV bias; Number of Parameters: 32.5B; Number of Paramaters (Non-Embedding): 31.0B; Number of Layers: 64; Number of Attention Heads (GQA): 40 for Q and 8 for KV; Context Length: 131,072 tokens

Description of the training:

DeepSeek-R1-Distill-Qwen-32B is derived from Qwen-2.5 series, which are originally licensed under Apache 2.0 License, and now finetuned with 800k samples curated with DeepSeek-R1.

License:

MIT License

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;temperature=0.6;until=["\nclass","\ndef","\n#","\nif","\nprint"]; \nrucodeeval - do_sample=true;temperature=0.6;until=["\nclass","\ndef","\n#","\nif","\nprint"];

The size of the context:
simplear, chegeka, rudetox, rumultiar, use, multiq, rumodar, ruhumaneval, rucodeeval - 3072

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:
{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<|User|>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<|Assistant|><|tool▁calls▁begin|><|tool▁call▁begin|>' + tool['type'] + '<|tool▁sep|>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<|tool▁call▁end|>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<|tool▁call▁begin|>' + tool['type'] + '<|tool▁sep|>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<|tool▁call▁end|>'}}{{'<|tool▁calls▁end|><|end▁of▁sentence|>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<|tool▁outputs▁end|>' + message['content'] + '<|end▁of▁sentence|>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<|Assistant|>' + content + '<|end▁of▁sentence|>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<|tool▁outputs▁begin|><|tool▁output▁begin|>' + message['content'] + '<|tool▁output▁end|>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<|tool▁output▁begin|>' + message['content'] + '<|tool▁output▁end|>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<|tool▁outputs▁end|>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<|Assistant|><think>\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
DeepSeek-R1-Distill-Qwen-32B
НГУ
0.367 0.4 0.8 0.2 0.267 0.6 0.167 - 0.033 0.033 0.1 0 0.133 0.1 0.1 0.433 0.1 0.067 0.133 0.067 0.1 0.733 0.467 0.233 0.333 0.7 0.233 0.333 0.333 0.3 0.433
Model, team Honest Helpful Harmless
DeepSeek-R1-Distill-Qwen-32B
НГУ
0.836 0.847 0.914
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-R1-Distill-Qwen-32B
НГУ
0.615 0.512 0.882 0.812 0.791 0.816 0.728 0.72 0.775 0.677 0.579 0.651 0.48 0.741 0.822 0.691 0.73 0.778 0.756 0.809 0.426 0.801 0.84 0.728 0.75 0.84 0.503 0.657 0.739 0.64 0.77 0.802 0.712 0.838 0.747 0.825 0.82 0.871 0.788 0.778 0.848 0.787 0.655 0.883 0.83 0.792 0.794 0.848 0.606 0.692 0.79 0.831 0.808 0.845 0.85 0.764 0.808
Model, team SIM FL STA
DeepSeek-R1-Distill-Qwen-32B
НГУ
0.719 0.625 0.72
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-R1-Distill-Qwen-32B
НГУ
0.6 0.842 0.717 0.722 0.842 0.828 0.621 0.702 0.808 0.815 0.795 0.733 0.55 0.767 0.766 0.741 0.729 0.8 0.754 0.825 0.825 0.915 0.8 0.84 0.844 0.773 0.756 0.596 0.93 0.889 0.889 0.846 0.714 0.842 0.742 0.821 0.867 0.8 0.86 0.862 0.861 0.73 0.867 1 0.879 0.844 0.914 0.818 0.738 0.877 0.822 0.754 0.823 0.727 0.651 0.807 0.856
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
DeepSeek-R1-Distill-Qwen-32B
НГУ
0.536 0.546 0.576 0.483 0.427
Model, team Virtue Law Moral Justice Utilitarianism
DeepSeek-R1-Distill-Qwen-32B
НГУ
0.544 0.522 0.569 0.477 0.45
Model, team Virtue Law Moral Justice Utilitarianism
DeepSeek-R1-Distill-Qwen-32B
НГУ
0.554 0.539 0.593 0.529 0.454
Model, team Women Men LGBT Nationalities Migrants Other
DeepSeek-R1-Distill-Qwen-32B
НГУ
0.889 0.771 0.941 0.892 1 0.869