Text Industrial Code Multi
Leaderboard Tasks How MERA works
Return to the leaderboard

Meta-Llama-3-70B-Instruct

MERA Created at 22.09.2024 20:46
0.528
The overall result
113
Place in the rating
Weak tasks:
313
RWSD
117
PARus
43
RCB
131
ruEthics
31
MultiQ
117
ruWorldTree
89
ruOpenBookQA
94
CheGeKa
94
ruMMLU
166
ruHateSpeech
77
ruDetox
76
ruHHH
98
ruTiE
245
ruHumanEval
113
USE
104
MathLogicQA
104
ruMultiAr
167
SimpleAr
137
LCS
71
BPS
250
ruModAr
86
MaMuRAMu
+18
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.14 Accuracy
RCB 0.587 / 0.493 Accuracy F1 macro
USE 0.281 Grade norm
RWSD 0.519 Accuracy
PARus 0.892 Accuracy
ruTiE 0.807 Accuracy
MultiQ 0.595 / 0.421 F1 Exact match
CheGeKa 0.327 / 0.269 F1 Exact match
ruModAr 0.475 Exact match
MaMuRAMu 0.803 Accuracy
ruMultiAr 0.343 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.534 Accuracy
ruWorldTree 0.958 / 0.958 Accuracy F1 macro
ruOpenBookQA 0.905 / 0.727 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 0.986 Accuracy
ruMMLU 0.71 Accuracy
SimpleAr 0.98 Exact match
ruHumanEval 0.012 / 0.012 / 0.012 Pass@k
ruHHH 0.848
ruHateSpeech 0.774
ruDetox 0.31
ruEthics
Correct God Ethical
Virtue 0.366 0.396 0.456
Law 0.356 0.374 0.439
Moral 0.403 0.423 0.488
Justice 0.331 0.348 0.408
Utilitarianism 0.307 0.346 0.38

Information about the submission

Mera version
v.1.2.0
Torch Version
2.4.0
The version of the codebase
9b26db97
CUDA version
12.1
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.44.2
The number of GPUs and their type
8 x NVIDIA H100 80GB HBM3
Architecture
vllm

Team:

MERA

Name of the ML model:

Meta-Llama-3-70B-Instruct

Link to the ML model:

https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct

Model size

70.6B

Model type:

Opened

SFT

Additional links:

https://ai.meta.com/research/publications/the-llama-3-herd-of-models/

Architecture description:

An auto-regressive language model that uses an optimized transformer architecture, with context length of 8K tokens. The model uses Grouped-Query Attention.

Description of the training:

Training Time is 6.4M GPU hours on hardware of type H100-80GB.

Pretrain data:

Llama 3 was pretrained on over 15 trillion tokens. The pretraining data includes data from publicly available sources and has a cutoff of December 2023. The fine-tuning data includes publicly available instruction datasets, as well as over 10M human-annotated examples.

License:

https://llama.meta.com/llama3/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;until=["\nclass","\ndef","\n#","\nif","\nprint"];temperature=0.6; \nrucodeeval - do_sample=true;until=["\nclass","\ndef","\n#","\nif","\nprint"];temperature=0.6;

The size of the context:
simplear, bps, lcs, chegeka, mathlogicqa, parus, rcb, rudetox, ruhatespeech, ruworldtree, ruopenbookqa, rumultiar, use, rwsd, mamuramu, rummlu, multiq, rumodar, ruethics, ruhhh, ruhumaneval, rucodeeval - 8192 \nrutie - 2000 \nrutie - 5000

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 loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|> \n \n'+ message['content'] | trim + '<|eot_id|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{% if add_generation_prompt %}{{ '<|start_header_id|>assistant<|end_header_id|> \n \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
Meta-Llama-3-70B-Instruct
MERA 		0.333 0.4 0.767 0.267 0.067 0.433 0.067 - 0.033 0.033 0.067 0 0.2 0.033 0 0.467 0.067 0 0.033 0.033 0.067 0.733 0.267 0.333 0.167 0.592 0.167 0.233 0.6 0.333 0.533
Model, team Honest Helpful Harmless
Meta-Llama-3-70B-Instruct
MERA 		0.836 0.797 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
Meta-Llama-3-70B-Instruct
MERA 		0.696 0.512 0.829 0.842 0.84 0.811 0.748 0.73 0.799 0.758 0.596 0.556 0.62 0.806 0.83 0.714 0.77 0.826 0.556 0.802 0.566 0.83 0.49 0.734 0.545 0.84 0.535 0.648 0.751 0.56 0.79 0.876 0.724 0.899 0.736 0.726 0.47 0.874 0.563 0.635 0.848 0.809 0.697 0.668 0.88 0.597 0.877 0.489 0.528 0.693 0.62 0.865 0.777 0.786 0.83 0.812 0.865
Model, team SIM FL STA
Meta-Llama-3-70B-Instruct
MERA 		0.602 0.695 0.784
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
Meta-Llama-3-70B-Instruct
MERA 		0.644 0.881 0.767 0.667 0.855 0.81 0.621 0.702 0.923 0.785 0.769 0.758 0.633 0.791 0.813 0.765 0.729 0.778 0.737 0.86 0.772 0.898 0.889 0.817 0.822 0.803 0.846 0.772 0.912 0.778 0.844 0.859 0.768 0.912 0.712 0.821 0.8 0.844 0.719 0.738 0.894 0.889 0.822 0.978 0.914 0.889 0.914 0.864 0.769 0.93 0.822 0.884 0.785 0.727 0.721 0.731 0.856
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3-70B-Instruct
MERA 		0.366 0.356 0.403 0.331 0.307
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3-70B-Instruct
MERA 		0.396 0.374 0.423 0.348 0.346
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3-70B-Instruct
MERA 		0.456 0.439 0.488 0.408 0.38
Model, team Women Men LGBT Nationalities Migrants Other
Meta-Llama-3-70B-Instruct
MERA 		0.769 0.629 0.647 0.811 0.714 0.885

MERA

Useful links

© 2025 MERA. All rights reserved.
User agreement Privacy policy

We use cookies to store data. By continuing to use the site, you consent to work with these files