Meta-Llama-3.1-8B

MERA Created at 20.09.2024 17:20
0.449
The overall result
241
Place in the rating
Weak tasks:
245
RWSD
306
PARus
265
RCB
381
ruEthics
321
MultiQ
304
ruWorldTree
319
ruOpenBookQA
181
CheGeKa
294
ruMMLU
379
ruHateSpeech
467
ruDetox
294
ruHHH
384
ruTiE
101
ruHumanEval
295
USE
400
MathLogicQA
303
ruMultiAr
261
SimpleAr
238
LCS
369
BPS
90
ruModAr
270
MaMuRAMu
129
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.116 Accuracy
RCB 0.527 / 0.455 Accuracy F1 macro
USE 0.132 Grade norm
RWSD 0.546 Accuracy
PARus 0.752 Accuracy
ruTiE 0.56 Accuracy
MultiQ 0.327 / 0.138 F1 Exact match
CheGeKa 0.177 / 0.137 F1 Exact match
ruModAr 0.71 Exact match
MaMuRAMu 0.641 Accuracy
ruMultiAr 0.242 Exact match
ruCodeEval 0.148 / 0.358 / 0.433 Pass@k
MathLogicQA 0.34 Accuracy
ruWorldTree 0.81 / 0.808 Accuracy F1 macro
ruOpenBookQA 0.69 / 0.687 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 0.842 Accuracy
ruMMLU 0.521 Accuracy
SimpleAr 0.949 Exact match
ruHumanEval 0.195 / 0.372 / 0.445 Pass@k
ruHHH 0.635
ruHateSpeech 0.604
ruDetox 0.081
ruEthics
Correct God Ethical
Virtue 0.193 0.198 0.164
Law 0.184 0.192 0.141
Moral 0.21 0.199 0.181
Justice 0.163 0.205 0.119
Utilitarianism 0.152 0.169 0.179

Information about the submission

Mera version
v.1.2.0
Torch Version
2.3.1
The version of the codebase
aec92f8
CUDA version
12.1
Precision of the model weights
float16
Seed
1234
Batch
1
Transformers version
4.43.3
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Meta-Llama-3.1-8B

Model size

8.0B

Model type:

Opened

Pretrain

Additional links:

https://ai.meta.com/blog/meta-llama-3-1/

Architecture description:

An auto-regressive language model that uses an optimized transformer architecture, with context length of 128k tokens. The model uses Grouped-Query Attention.

Description of the training:

Training Time is 1.46M GPU hours on hardware of type H100-80GB.

Pretrain data:

Llama 3.1 was pretrained on over 15 trillion tokens. The pretraining data includes data from publicly available sources and 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 - 131072

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-3.1-8B
MERA
0.333 0.167 0.567 0.133 0.033 0.133 0 - 0.033 0.033 0 0 0.033 0.033 0.167 0.317 0.067 0 0.067 0 0 0.3 0.067 0.033 0.033 0.25 0.033 0.067 0.2 0.1 0.233
Model, team Honest Helpful Harmless
Meta-Llama-3.1-8B
MERA
0.607 0.644 0.655
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-3.1-8B
MERA
0.467 0.482 0.638 0.688 0.614 0.751 0.709 0.588 0.552 0.556 0.386 0.421 0.35 0.602 0.625 0.581 0.6 0.576 0.322 0.618 0.283 0.667 0.27 0.48 0.393 0.62 0.37 0.602 0.673 0.37 0.7 0.769 0.528 0.747 0.558 0.462 0.35 0.652 0.404 0.468 0.662 0.471 0.552 0.422 0.646 0.463 0.706 0.396 0.369 0.52 0.46 0.717 0.526 0.542 0.64 0.715 0.642
Model, team SIM FL STA
Meta-Llama-3.1-8B
MERA
0.863 0.582 0.181
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-3.1-8B
MERA
0.467 0.723 0.583 0.574 0.697 0.724 0.552 0.632 0.75 0.585 0.641 0.608 0.483 0.651 0.673 0.531 0.701 0.622 0.351 0.702 0.351 0.729 0.533 0.675 0.6 0.606 0.628 0.596 0.842 0.689 0.733 0.744 0.643 0.807 0.561 0.589 0.489 0.689 0.439 0.662 0.714 0.667 0.733 0.644 0.845 0.8 0.759 0.614 0.615 0.807 0.733 0.638 0.747 0.571 0.628 0.462 0.756
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3.1-8B
MERA
0.193 0.184 0.21 0.163 0.152
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3.1-8B
MERA
0.198 0.192 0.199 0.205 0.169
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3.1-8B
MERA
0.164 0.141 0.181 0.119 0.179
Model, team Women Men LGBT Nationalities Migrants Other
Meta-Llama-3.1-8B
MERA
0.639 0.514 0.412 0.757 0.143 0.607