LFM2-350M

MERA Created at 26.03.2026 22:05
0.215
The overall result
572
Place in the rating
Weak tasks:
606
RWSD
583
PARus
584
RCB
601
MultiQ
546
ruWorldTree
546
ruOpenBookQA
608
CheGeKa
549
ruMMLU
562
ruHateSpeech
622
ruDetox
594
ruHHH
557
ruTiE
583
USE
590
MathLogicQA
450
ruMultiAr
550
SimpleAr
541
LCS
508
BPS
567
ruModAr
553
MaMuRAMu
+16
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.048 Accuracy
RCB 0.333 / 0.33 Accuracy F1 macro
USE 0.027 Grade norm
RWSD 0.335 Accuracy
PARus 0.48 Accuracy
ruTiE 0.499 Accuracy
MultiQ 0.062 / 0.017 F1 Exact match
CheGeKa 0.003 / 0 F1 Exact match
ruModAr 0.001 Exact match
MaMuRAMu 0.326 Accuracy
ruMultiAr 0.133 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.234 Accuracy
ruWorldTree 0.396 / 0.391 Accuracy F1 macro
ruOpenBookQA 0.388 / 0.367 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 0.547 Accuracy
ruMMLU 0.307 Accuracy
SimpleAr 0.13 Exact match
ruHumanEval 0 / 0 / 0 Pass@k
ruHHH 0.399
ruHateSpeech 0.475
ruDetox 0.037
ruEthics
Correct God Ethical
Virtue -0.013 0.041 0.06
Law -0.024 0.037 0.085
Moral -0.023 0.017 0.047
Justice -0.02 0.011 0.068
Utilitarianism -0.018 0.026 0.032

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
1 x NVIDIA A100-SXM4-80GB
Architecture
vllm

Team:

MERA

Name of the ML model:

LFM2-350M

Model size

354.5M

Model type:

Opened

SFT

Additional links:

https://docs.liquid.ai/lfm/getting-started/welcome

Architecture description:

Hybrid model with multiplicative gates and short convolutions: 10 double-gated short-range LIV convolution blocks and 6 grouped query attention (GQA) blocks.

Description of the training:

Training approach: Knowledge distillation using LFM1-7B as teacher model Very large-scale SFT on 50% downstream tasks, 50% general domains Custom DPO with length normalization and semi-online datasets Iterative model merging

Pretrain data:

Approximately 75% English, 20% multilingual, and 5% code data sourced from the web and licensed materials.

License:

lfm1.0

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:
32768

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 You are Olmo, a helpful AI assistant built by Ai2. Your date cutoff is December 2024, and your model weights are available at https://huggingface.co/allenai. ' -}}{%- if tools is none or (tools | length) == 0 -%}{{- 'You do not currently have access to any functions. <functions></functions><|im_end|> ' -}}{%- 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|> ' -}}{%- endif -%}{%- endif -%}{%- for message in messages -%}{%- if message['role'] == 'system' -%}{{- '<|im_start|>system ' + 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|> ' -}}{%- elif message['role'] == 'user' -%}{{- '<|im_start|>user ' + message['content'] + '<|im_end|> ' -}}{%- elif message['role'] == 'assistant' -%}{{- '<|im_start|>assistant ' -}}{%- 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 -%}{{ ' ' }}{%- endif -%}{% else %}{{- tool_call -}}{%- endif %}{%- endfor %}{{- '</function_calls>' -}}{%- endif -%}{%- if not loop.last -%}{{- '<|im_end|>' + ' ' -}}{%- else -%}{{- eos_token -}}{%- endif -%}{%- elif message['role'] == 'environment' -%}{{- '<|im_start|>environment ' + message['content'] + '<|im_end|> ' -}}{%- elif message['role'] == 'tool' -%}{{- '<|im_start|>environment ' + message['content'] + '<|im_end|> ' -}}{%- 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
LFM2-350M
MERA
0 0 0.133 0.1 0 0 0 - 0 0 0 0 0 0 0 0.033 0 0.033 0 0 0 0 0.033 0 0 0.033 0.067 0.067 0.133 0.1 0.067
Model, team Honest Helpful Harmless
LFM2-350M
MERA
0.426 0.356 0.414
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
LFM2-350M
MERA
0.274 0.307 0.382 0.423 0.402 0.353 0.301 0.347 0.265 0.35 0.184 0.262 0.3 0.435 0.292 0.379 0.38 0.264 0.256 0.252 0.24 0.316 0.21 0.26 0.152 0.33 0.274 0.343 0.269 0.34 0.38 0.537 0.319 0.455 0.328 0.295 0.29 0.368 0.278 0.305 0.338 0.246 0.31 0.257 0.328 0.352 0.328 0.211 0.234 0.318 0.33 0.35 0.323 0.366 0.33 0.376 0.326
Model, team SIM FL STA
LFM2-350M
MERA
0.425 0.306 0.383
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
LFM2-350M
MERA
0.378 0.307 0.267 0.38 0.316 0.31 0.414 0.281 0.231 0.415 0.346 0.308 0.25 0.326 0.322 0.296 0.374 0.444 0.228 0.298 0.158 0.254 0.111 0.314 0.222 0.379 0.372 0.281 0.561 0.267 0.556 0.231 0.366 0.281 0.303 0.304 0.333 0.267 0.298 0.385 0.327 0.349 0.4 0.311 0.569 0.644 0.259 0.159 0.354 0.368 0.4 0.304 0.456 0.286 0.233 0.24 0.344
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
LFM2-350M
MERA
-0.013 -0.024 -0.023 -0.02 -0.018
Model, team Virtue Law Moral Justice Utilitarianism
LFM2-350M
MERA
0.041 0.037 0.017 0.011 0.026
Model, team Virtue Law Moral Justice Utilitarianism
LFM2-350M
MERA
0.06 0.085 0.047 0.068 0.032
Model, team Women Men LGBT Nationalities Migrants Other
LFM2-350M
MERA
0.444 0.429 0.471 0.459 0.429 0.574