Olmo-3.1-32B-Instruct-DPO

MERA Created at 18.03.2026 08:25
0.468
The overall result
201
Place in the rating
Weak tasks:
322
RWSD
279
PARus
251
RCB
173
ruEthics
151
MultiQ
274
ruWorldTree
238
ruOpenBookQA
282
CheGeKa
281
ruMMLU
297
ruHateSpeech
169
ruDetox
235
ruHHH
177
ruTiE
202
ruHumanEval
163
USE
117
MathLogicQA
132
ruMultiAr
86
SimpleAr
46
LCS
352
BPS
212
ruModAr
345
MaMuRAMu
197
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.2 Accuracy
RCB 0.532 / 0.508 Accuracy F1 macro
USE 0.209 Grade norm
RWSD 0.519 Accuracy
PARus 0.778 Accuracy
ruTiE 0.734 Accuracy
MultiQ 0.458 / 0.307 F1 Exact match
CheGeKa 0.105 / 0.07 F1 Exact match
ruModAr 0.5 Exact match
MaMuRAMu 0.563 Accuracy
ruMultiAr 0.332 Exact match
ruCodeEval 0.043 / 0.123 / 0.165 Pass@k
MathLogicQA 0.518 Accuracy
ruWorldTree 0.85 / 0.85 Accuracy F1 macro
ruOpenBookQA 0.793 / 0.635 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 0.859 Accuracy
ruMMLU 0.529 Accuracy
SimpleAr 0.992 Exact match
ruHumanEval 0.054 / 0.147 / 0.195 Pass@k
ruHHH 0.697
ruHateSpeech 0.672
ruDetox 0.222
ruEthics
Correct God Ethical
Virtue 0.342 0.355 0.454
Law 0.325 0.33 0.418
Moral 0.356 0.364 0.473
Justice 0.287 0.291 0.401
Utilitarianism 0.286 0.31 0.39

Information about the submission

Mera version
v1.2.0
Torch Version
2.10.0
The version of the codebase
0ac3a14
CUDA version
12.8
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.57.6
The number of GPUs and their type
4 x NVIDIA A100-SXM4-80GB
Architecture
vllm

Team:

MERA

Name of the ML model:

Olmo-3.1-32B-Instruct-DPO

Model size

32.0B

Model type:

Opened

SFT

Additional links:

https://arxiv.org/abs/2512.13961

Architecture description:

Model Description Developed by: Allen Institute for AI (Ai2) Model type: a Transformer style autoregressive language model. Language(s) (NLP): English

Description of the training:

Model Details Stage 1: SFT supervised fine-tuning on the Dolci-Think-SFT-7B dataset. This dataset consits of math, code, chat, and general knowledge queries. Datasets: Dolci-Think-SFT-7B, Dolci-Instruct-SFT Stage 2:DPO direct preference optimization on the Dolci-Think-DPO-7B dataset. This dataset consits of math, code, chat, and general knowledge queries. Datasets: Dolci-Think-DPO-7B, Dolci-Instruct-DPO

Pretrain data:

Date cutoff: Dec. 2024

License:

Apache 2.0

Inference parameters

Generation Parameters:
simplear - 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"]; \nrudetox - do_sample=false;until=["\n"]; \nchegeka - 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"];

The size of the context:
simplear, rumultiar, use, multiq, rumodar, rudetox, chegeka, rucodeeval, ruhumaneval - 65536

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:
{%- set has_system = messages|selectattr('role', 'equalto', 'system')|list|length > 0 -%}{%- if not has_system -%}{{- '<|im_start|>system \nYou are a helpful function-calling AI assistant. ' -}}{%- if tools is none or (tools | length) == 0 -%}{{- 'You do not currently have access to any functions. <functions></functions><|im_end|> \n' -}}{%- else -%}{{- 'You are provided with function signatures within <functions></functions> XML tags. You may call one or more functions to assist with the user query. Output any function calls within <function_calls></function_calls> XML tags. Do not make assumptions about what values to plug into functions.' -}}{{- '<functions>' -}}{{- tools | tojson -}}{{- '</functions><|im_end|> \n' -}}{%- endif -%}{%- endif -%}{%- for message in messages -%}{%- if message['role'] == 'system' -%}{{- '<|im_start|>system \n' + message['content'] -}}{%- if tools is not none -%}{{- '<functions>' -}}{{- tools | tojson -}}{{- '</functions>' -}}{%- elif message.get('functions', none) is not none -%}{{- ' <functions>' + message['functions'] + '</functions>' -}}{%- endif -%}{{- '<|im_end|> \n' -}}{%- elif message['role'] == 'user' -%}{{- '<|im_start|>user \n' + message['content'] + '<|im_end|> \n' -}}{%- elif message['role'] == 'assistant' -%}{{- '<|im_start|>assistant \n' -}}{%- if message.get('content', none) is not none -%}{{- message['content'] -}}{%- endif -%}{%- if message.get('function_calls', none) is not none -%}{{- '<function_calls>' + message['function_calls'] + '</function_calls>' -}}{% elif message.get('tool_calls', none) is not none %}{{- '<function_calls>' -}}{%- for tool_call in message['tool_calls'] %}{%- if tool_call is mapping and tool_call.get('function', none) is not none %}{%- set args = tool_call['function']['arguments'] -%}{%- set ns = namespace(arguments_list=[]) -%}{%- for key, value in args.items() -%}{%- set ns.arguments_list = ns.arguments_list + [key ~ '=' ~ (value | tojson)] -%}{%- endfor -%}{%- set arguments = ns.arguments_list | join(', ') -%}{{- tool_call['function']['name'] + '(' + arguments + ')' -}}{%- if not loop.last -%}{{ ' \n' }}{%- endif -%}{% else %}{{- tool_call -}}{%- endif %}{%- endfor %}{{- '</function_calls>' -}}{%- endif -%}{%- if not loop.last -%}{{- '<|im_end|>' + ' \n' -}}{%- else -%}{{- eos_token -}}{%- endif -%}{%- elif message['role'] == 'environment' -%}{{- '<|im_start|>environment \n' + message['content'] + '<|im_end|> \n' -}}{%- elif message['role'] == 'tool' -%}{{- '<|im_start|>environment \n' + message['content'] + '<|im_end|> \n' -}}{%- endif -%}{%- if loop.last and add_generation_prompt -%}{{- '<|im_start|>assistant \n' -}}{%- endif -%}{%- endfor -%}

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
Olmo-3.1-32B-Instruct-DPO
MERA
0.267 0.367 0.8 0.267 0 0.267 0.033 - 0 0.1 0 0 0.2 0.1 0.2 0.333 0.033 0 0.033 0 0.033 0.4 0.267 0.1 0.033 0.417 0.133 0.133 0.3 0.333 0.367
Model, team Honest Helpful Harmless
Olmo-3.1-32B-Instruct-DPO
MERA
0.656 0.729 0.707
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
Olmo-3.1-32B-Instruct-DPO
MERA
0.43 0.494 0.539 0.735 0.605 0.682 0.641 0.563 0.522 0.61 0.412 0.516 0.43 0.611 0.535 0.59 0.63 0.465 0.533 0.511 0.375 0.491 0.31 0.474 0.518 0.48 0.395 0.546 0.612 0.39 0.64 0.752 0.595 0.657 0.506 0.594 0.43 0.632 0.497 0.537 0.641 0.39 0.517 0.618 0.65 0.523 0.578 0.489 0.418 0.508 0.55 0.717 0.531 0.576 0.85 0.63 0.591
Model, team SIM FL STA
Olmo-3.1-32B-Instruct-DPO
MERA
0.462 0.66 0.775
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
Olmo-3.1-32B-Instruct-DPO
MERA
0.356 0.525 0.55 0.574 0.5 0.603 0.517 0.474 0.346 0.492 0.603 0.608 0.342 0.581 0.409 0.444 0.636 0.467 0.456 0.491 0.491 0.661 0.689 0.568 0.778 0.47 0.667 0.421 0.825 0.578 0.822 0.603 0.571 0.632 0.515 0.625 0.644 0.667 0.474 0.569 0.571 0.381 0.8 0.8 0.828 0.844 0.586 0.773 0.662 0.737 0.933 0.319 0.696 0.597 0.698 0.316 0.589
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Olmo-3.1-32B-Instruct-DPO
MERA
0.342 0.325 0.356 0.287 0.286
Model, team Virtue Law Moral Justice Utilitarianism
Olmo-3.1-32B-Instruct-DPO
MERA
0.355 0.33 0.364 0.291 0.31
Model, team Virtue Law Moral Justice Utilitarianism
Olmo-3.1-32B-Instruct-DPO
MERA
0.454 0.418 0.473 0.401 0.39
Model, team Women Men LGBT Nationalities Migrants Other
Olmo-3.1-32B-Instruct-DPO
MERA
0.722 0.6 0.647 0.595 0.429 0.705