Meta-Llama-3.1-405B-Instruct

MERA Created at 22.09.2024 20:50
0.638
The overall result
47
Place in the rating
Weak tasks:
38
RWSD
61
PARus
20
RCB
25
ruEthics
92
MultiQ
59
ruWorldTree
40
ruOpenBookQA
28
CheGeKa
33
ruMMLU
26
ruHateSpeech
52
ruDetox
44
ruHHH
31
ruTiE
45
ruHumanEval
49
USE
59
MathLogicQA
79
ruMultiAr
41
SimpleAr
42
LCS
55
BPS
79
ruModAr
34
MaMuRAMu
36
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.214 Accuracy
RCB 0.6 / 0.554 Accuracy F1 macro
USE 0.379 Grade norm
RWSD 0.685 Accuracy
PARus 0.922 Accuracy
ruTiE 0.88 Accuracy
MultiQ 0.524 / 0.329 F1 Exact match
CheGeKa 0.471 / 0.351 F1 Exact match
ruModAr 0.727 Exact match
MaMuRAMu 0.861 Accuracy
ruMultiAr 0.395 Exact match
ruCodeEval 0.458 / 0.47 / 0.476 Pass@k
MathLogicQA 0.708 Accuracy
ruWorldTree 0.981 / 0.981 Accuracy F1 macro
ruOpenBookQA 0.938 / 0.938 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 0.991 Accuracy
ruMMLU 0.808 Accuracy
SimpleAr 0.997 Exact match
ruHumanEval 0.408 / 0.425 / 0.427 Pass@k
ruHHH 0.865
ruHateSpeech 0.872
ruDetox 0.332
ruEthics
Correct God Ethical
Virtue 0.497 0.432 0.602
Law 0.488 0.419 0.579
Moral 0.542 0.462 0.64
Justice 0.457 0.392 0.544
Utilitarianism 0.432 0.387 0.528

Information about the submission

Mera version
v.1.2.0
Torch Version
-
The version of the codebase
1aa3a9aa
CUDA version
-
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.43.4
The number of GPUs and their type
16 x NVIDIA H100 80GB HBM3
Architecture
openai-completions

Team:

MERA

Name of the ML model:

Meta-Llama-3.1-405B-Instruct

Model size

405.0B

Model type:

Opened

SFT

Architecture description:

A pretrained and instruction tuned 405B auto-regressive generative model with Grouped-Query Attention. Optimized for multilingual dialogue use cases.

Description of the training:

The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.

Pretrain data:

Llama 3.1 was pretrained on ~15 trillion tokens of data from publicly available sources. The fine-tuning data includes publicly available instruction datasets, as well as over 25M synthetically generated examples. The pretraining data has a cutoff of December 2023.

License:

https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/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, rummlu, ruworldtree, ruopenbookqa, rumultiar, use, rwsd, mamuramu, multiq, rumodar, ruethics, ruhhh, rutie, ruhumaneval, rucodeeval - 2048

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-3.1-405B-Instruct
MERA
0.667 0.6 0.9 0.133 0.167 0.667 0.167 - 0.033 0 0.033 0 0.333 0.033 0.1 0.517 0.067 0.067 0.1 0.133 0.133 0.733 0.533 0.333 0.233 0.75 0.367 0.233 0.567 0.8 0.733
Model, team Honest Helpful Harmless
Meta-Llama-3.1-405B-Instruct
MERA
0.869 0.847 0.879
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-3.1-405B-Instruct
MERA
0.77 0.536 0.914 0.91 0.912 0.9 0.874 0.839 0.867 0.812 0.702 0.667 0.69 0.861 0.916 0.803 0.83 0.965 0.689 0.84 0.754 0.889 0.64 0.746 0.714 0.9 0.656 0.787 0.808 0.61 0.84 0.909 0.834 0.949 0.808 0.833 0.54 0.916 0.709 0.793 0.914 0.915 0.717 0.857 0.929 0.773 0.917 0.663 0.652 0.8 0.76 0.924 0.854 0.899 0.93 0.873 0.922
Model, team SIM FL STA
Meta-Llama-3.1-405B-Instruct
MERA
0.809 0.666 0.633
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-3.1-405B-Instruct
MERA
0.756 0.911 0.833 0.741 0.921 0.879 0.741 0.807 0.962 0.815 0.821 0.817 0.583 0.899 0.854 0.765 0.757 0.844 0.877 0.895 0.912 0.932 0.844 0.899 0.844 0.894 0.872 0.86 0.93 0.867 0.889 0.936 0.875 0.93 0.773 0.857 0.867 0.867 0.772 0.846 0.939 0.952 0.844 0.978 0.966 0.911 0.948 0.909 0.815 0.912 0.911 0.971 0.861 0.753 0.767 0.889 0.922
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3.1-405B-Instruct
MERA
0.497 0.488 0.542 0.457 0.432
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3.1-405B-Instruct
MERA
0.432 0.419 0.462 0.392 0.387
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3.1-405B-Instruct
MERA
0.602 0.579 0.64 0.544 0.528
Model, team Women Men LGBT Nationalities Migrants Other
Meta-Llama-3.1-405B-Instruct
MERA
0.88 0.771 0.882 0.892 0.714 0.918