LFM2-8B-A1B

MERA Created at 03.04.2026 13:15
0.278
The overall result
516
Place in the rating
Weak tasks:
595
RWSD
507
PARus
441
RCB
502
ruEthics
484
MultiQ
454
ruWorldTree
459
ruOpenBookQA
540
CheGeKa
464
ruMMLU
539
ruHateSpeech
475
ruDetox
567
ruHHH
346
ruTiE
276
ruHumanEval
356
USE
593
MathLogicQA
347
ruMultiAr
265
SimpleAr
203
LCS
409
BPS
551
ruModAr
467
MaMuRAMu
252
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.126 Accuracy
RCB 0.454 / 0.281 Accuracy F1 macro
USE 0.098 Grade norm
RWSD 0.373 Accuracy
PARus 0.518 Accuracy
ruTiE 0.59 Accuracy
MultiQ 0.202 / 0.143 F1 Exact match
CheGeKa 0.009 / 0 F1 Exact match
ruModAr 0.006 Exact match
MaMuRAMu 0.438 Accuracy
ruMultiAr 0.212 Exact match
ruCodeEval 0.004 / 0.018 / 0.037 Pass@k
MathLogicQA 0.198 Accuracy
ruWorldTree 0.621 / 0.501 Accuracy F1 macro
ruOpenBookQA 0.538 / 0.431 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 0.768 Accuracy
ruMMLU 0.385 Accuracy
SimpleAr 0.949 Exact match
ruHumanEval 0.009 / 0.033 / 0.049 Pass@k
ruHHH 0.461
ruHateSpeech 0.494
ruDetox 0.079
ruEthics
Correct God Ethical
Virtue 0.081 0.046 0.187
Law 0.054 0.019 0.178
Moral 0.075 0.024 0.194
Justice 0.057 0.015 0.175
Utilitarianism 0.051 0.063 0.186

Information about the submission

Mera version
v1.2.0
Torch Version
2.9.0
The version of the codebase
2a537dd
CUDA version
12.8
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
5.5.0.dev0
The number of GPUs and their type
1 x NVIDIA A100-SXM4-80GB
Architecture
hf

Team:

MERA

Name of the ML model:

LFM2-8B-A1B

Model size

8.0B

Model type:

Opened

SFT

MoE

Additional links:

https://arxiv.org/abs/2511.23404

Architecture description:

LFM2 is a new generation of hybrid models developed by Liquid AI, specifically designed for edge AI and on-device deployment. It sets a new standard in terms of quality, speed, and memory efficiency.

Description of the training:

Hybrid model with multiplicative gates and short convolutions: 18 double-gated short-range LIV convolution blocks and 6 grouped query attention (GQA) blocks.

Pretrain data:

The SFT corpus we use comprises approximately 5.39 million samples for the 350M, 700M, 1.2B, and 2.6B models, and 9.24 million samples for the 8B-A1B model (which includes additional math and code-focused data). The two datasets integrate 67 and 79 carefully curated sources, respectively, spanning open-source datasets, licensed data, and targeted synthetic generations optimized for on-device deployment scenarios, with multilingual samples integrated throughout each category.

License:

LFM Open License v1.0

Inference parameters

Generation Parameters:
rucodeeval - 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"];\nrutie_gen - do_sample=false;until=["\n"];\nbps_gen - do_sample=false;until=["\n"];\nchegeka - do_sample=false;until=["\n"];\nlcs_gen - do_sample=false;until=["\n"];\nmamuramu_gen - do_sample=false;until=["\n"];\nmathlogicqa_gen - do_sample=false;until=["\n"];\nmultiq - do_sample=false;until=["\n"];\nparus_gen - do_sample=false;until=["\n"];\nrcb_gen - do_sample=false;until=["\n"];\nrudetox - do_sample=false;until=["\n"];\nruethics_gen - do_sample=false;until=["\n"];\nruhatespeech_gen - do_sample=false;until=["\n"];\nruhhh_gen - do_sample=false;until=["\n"];\nrummlu_gen - do_sample=false;until=["\n"];\nrumodar - do_sample=false;until=["\n"];\nrumultiar - do_sample=false;until=["\n"];\nruopenbookqa_gen - do_sample=false;until=["\n"];\nruworldtree_gen - do_sample=false;until=["\n"];\nrwsd_gen - do_sample=false;until=["\n"];\nsimplear - do_sample=false;until=["\n"];\nuse - do_sample=false;until=["\n", "."];

The size of the context:
128000

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:
{{- bos_token -}}{%- set system_prompt = \"\" -%}{%- set ns = namespace(system_prompt=\"\") -%}{%- if messages[0][\"role\"] == \"system\" -%} {%- set ns.system_prompt = messages[0][\"content\"] -%} {%- set messages = messages[1:] -%}{%- endif -%}{%- if tools -%} {%- set ns.system_prompt = ns.system_prompt + (\"\n\" if ns.system_prompt else \"\") + \"List of tools: <|tool_list_start|>[\" -%} {%- for tool in tools -%} {%- if tool is not string -%} {%- set tool = tool | tojson -%} {%- endif -%} {%- set ns.system_prompt = ns.system_prompt + tool -%} {%- if not loop.last -%} {%- set ns.system_prompt = ns.system_prompt + \", \" -%} {%- endif -%} {%- endfor -%} {%- set ns.system_prompt = ns.system_prompt + \"]<|tool_list_end|>\" -%}{%- endif -%}{%- if ns.system_prompt -%} {{- \"<|im_start|>system\n\" + ns.system_prompt + \"<|im_end|>\n\" -}}{%- endif -%}{%- for message in messages -%} {{- \"<|im_start|>\" + message[\"role\"] + \"\n\" -}} {%- set content = message[\"content\"] -%} {%- if content is not string -%} {%- set content = content | tojson -%} {%- endif -%} {%- if message[\"role\"] == \"tool\" -%} {%- set content = \"<|tool_response_start|>\" + content + \"<|tool_response_end|>\" -%} {%- endif -%} {{- content + \"<|im_end|>\n\" -}}{%- endfor -%}{%- if add_generation_prompt -%} {{- \"<|im_start|>assistant\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
LFM2-8B-A1B
MERA
0.267 0 0.5 0.233 0.033 0.067 0 - 0 0.133 0 0.067 0 0 0.133 0.233 0.033 0.033 0 0.033 0 0.067 0.067 0 0.033 0.1 0.033 0.167 0.2 0.167 0.2
Model, team Honest Helpful Harmless
LFM2-8B-A1B
MERA
0.426 0.407 0.552
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
LFM2-8B-A1B
MERA
0.393 0.44 0.467 0.581 0.513 0.557 0.544 0.447 0.404 0.381 0.289 0.31 0.26 0.519 0.423 0.434 0.47 0.354 0.156 0.397 0.236 0.427 0.22 0.353 0.268 0.49 0.311 0.509 0.473 0.27 0.46 0.661 0.442 0.556 0.442 0.41 0.21 0.51 0.265 0.369 0.51 0.301 0.421 0.223 0.496 0.343 0.456 0.096 0.238 0.408 0.29 0.515 0.367 0.391 0.52 0.473 0.456
Model, team SIM FL STA
LFM2-8B-A1B
MERA
0.443 0.503 0.534
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
LFM2-8B-A1B
MERA
0.333 0.465 0.417 0.37 0.368 0.569 0.414 0.368 0.5 0.446 0.462 0.458 0.308 0.357 0.374 0.358 0.514 0.511 0.316 0.526 0.211 0.424 0.378 0.462 0.4 0.545 0.41 0.421 0.737 0.4 0.644 0.487 0.42 0.474 0.364 0.589 0.4 0.489 0.316 0.369 0.437 0.46 0.6 0.333 0.638 0.689 0.466 0.523 0.569 0.561 0.578 0.377 0.557 0.377 0.163 0.327 0.511
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
LFM2-8B-A1B
MERA
0.081 0.054 0.075 0.057 0.051
Model, team Virtue Law Moral Justice Utilitarianism
LFM2-8B-A1B
MERA
0.046 0.019 0.024 0.015 0.063
Model, team Virtue Law Moral Justice Utilitarianism
LFM2-8B-A1B
MERA
0.187 0.178 0.194 0.175 0.186
Model, team Women Men LGBT Nationalities Migrants Other
LFM2-8B-A1B
MERA
0.454 0.486 0.294 0.568 0 0.639