Meta-Llama-3.1-8B-Instruct

MERA Created at 22.09.2024 20:54
0.491
The overall result
162
Place in the rating
Weak tasks:
192
RWSD
203
PARus
171
RCB
354
ruEthics
354
MultiQ
272
ruWorldTree
247
ruOpenBookQA
237
CheGeKa
268
ruMMLU
273
ruHateSpeech
197
ruDetox
217
ruHHH
240
ruTiE
47
ruHumanEval
274
USE
382
MathLogicQA
277
ruMultiAr
314
SimpleAr
249
LCS
114
BPS
114
ruModAr
264
MaMuRAMu
43
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.114 Accuracy
RCB 0.55 / 0.477 Accuracy F1 macro
USE 0.147 Grade norm
RWSD 0.562 Accuracy
PARus 0.846 Accuracy
ruTiE 0.684 Accuracy
MultiQ 0.303 / 0.183 F1 Exact match
CheGeKa 0.133 / 0.101 F1 Exact match
ruModAr 0.666 Exact match
MaMuRAMu 0.652 Accuracy
ruMultiAr 0.253 Exact match
ruCodeEval 0.412 / 0.619 / 0.689 Pass@k
MathLogicQA 0.353 Accuracy
ruWorldTree 0.853 / 0.853 Accuracy F1 macro
ruOpenBookQA 0.785 / 0.785 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 0.975 Accuracy
ruMMLU 0.545 Accuracy
SimpleAr 0.928 Exact match
ruHumanEval 0.402 / 0.602 / 0.659 Pass@k
ruHHH 0.713
ruHateSpeech 0.691
ruDetox 0.206
ruEthics
Correct God Ethical
Virtue 0.195 0.281 0.255
Law 0.199 0.277 0.255
Moral 0.232 0.301 0.272
Justice 0.141 0.238 0.23
Utilitarianism 0.188 0.275 0.248

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
bfloat16
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-Instruct

Model size

8.0B

Model type:

Opened

SFT

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. The fine-tuning data includes publicly available instruction datasets, as well as over 25M synthetically generated examples.

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-Instruct
MERA
0.3 0.1 0.567 0.1 0 0.2 0 - 0.067 0.067 0.033 0.033 0 0 0.067 0.333 0.033 0.033 0.033 0.067 0 0.5 0.133 0.067 0.033 0.267 0.033 0.033 0.3 0.2 0.267
Model, team Honest Helpful Harmless
Meta-Llama-3.1-8B-Instruct
MERA
0.656 0.678 0.81
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-Instruct
MERA
0.496 0.428 0.632 0.726 0.644 0.697 0.66 0.595 0.574 0.529 0.421 0.437 0.35 0.676 0.632 0.601 0.67 0.604 0.367 0.641 0.358 0.69 0.37 0.566 0.42 0.62 0.41 0.556 0.71 0.42 0.71 0.777 0.546 0.707 0.574 0.5 0.34 0.706 0.424 0.527 0.692 0.533 0.559 0.43 0.682 0.505 0.686 0.385 0.394 0.511 0.47 0.713 0.541 0.592 0.69 0.745 0.648
Model, team SIM FL STA
Meta-Llama-3.1-8B-Instruct
MERA
0.778 0.608 0.489
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-Instruct
MERA
0.556 0.723 0.617 0.611 0.632 0.741 0.534 0.684 0.654 0.569 0.667 0.617 0.467 0.636 0.626 0.58 0.692 0.644 0.439 0.772 0.298 0.712 0.467 0.704 0.667 0.712 0.641 0.596 0.86 0.644 0.756 0.705 0.688 0.807 0.53 0.643 0.511 0.689 0.561 0.6 0.788 0.667 0.667 0.622 0.862 0.844 0.759 0.75 0.677 0.825 0.711 0.652 0.658 0.623 0.605 0.497 0.711
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3.1-8B-Instruct
MERA
0.195 0.199 0.232 0.141 0.188
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3.1-8B-Instruct
MERA
0.281 0.277 0.301 0.238 0.275
Model, team Virtue Law Moral Justice Utilitarianism
Meta-Llama-3.1-8B-Instruct
MERA
0.255 0.255 0.272 0.23 0.248
Model, team Women Men LGBT Nationalities Migrants Other
Meta-Llama-3.1-8B-Instruct
MERA
0.667 0.657 0.706 0.703 0.429 0.77