Llama-2-13b-hf

MERA Created at 20.09.2024 15:32
0.345
The overall result
410
Place in the rating
Weak tasks:
223
RWSD
562
PARus
501
RCB
500
ruEthics
442
MultiQ
450
ruWorldTree
457
ruOpenBookQA
185
CheGeKa
429
ruMMLU
478
ruHateSpeech
562
ruDetox
494
ruHHH
492
ruTiE
164
ruHumanEval
456
USE
499
MathLogicQA
441
ruMultiAr
380
SimpleAr
407
LCS
469
BPS
235
ruModAr
403
MaMuRAMu
161
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.086 Accuracy
RCB 0.39 / 0.383 Accuracy F1 macro
USE 0.066 Grade norm
RWSD 0.554 Accuracy
PARus 0.49 Accuracy
ruTiE 0.507 Accuracy
MultiQ 0.238 / 0.126 F1 Exact match
CheGeKa 0.174 / 0.125 F1 Exact match
ruModAr 0.484 Exact match
MaMuRAMu 0.506 Accuracy
ruMultiAr 0.138 Exact match
ruCodeEval 0.087 / 0.196 / 0.244 Pass@k
MathLogicQA 0.293 Accuracy
ruWorldTree 0.623 / 0.624 Accuracy F1 macro
ruOpenBookQA 0.538 / 0.538 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 0.62 Accuracy
ruMMLU 0.411 Accuracy
SimpleAr 0.876 Exact match
ruHumanEval 0.102 / 0.197 / 0.238 Pass@k
ruHHH 0.489
ruHateSpeech 0.528
ruDetox 0.053
ruEthics
Correct God Ethical
Virtue 0.057 0.061 0.041
Law 0.059 0.064 0.055
Moral 0.064 0.066 0.064
Justice 0.068 0.074 0.04
Utilitarianism 0.017 0.033 0.018

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
4.42.3
The number of GPUs and their type
2 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Llama-2-13b-hf

Model size

13.0B

Model type:

Opened

Pretrain

Additional links:

https://arxiv.org/pdf/2307.09288

Architecture description:

Auto-regressive language model that uses an optimized transformer architecture, with context length of 4K tokens.

Description of the training:

Pretrained generative text model. The global batch-size is 4M tokens. The model was trained between January 2023 and July 2023.

Pretrain data:

A new mix of publicly available online data, the pretraining data volume is 2.0T.

License:

https://ai.meta.com/resources/models-and-libraries/llama-downloads/

Inference parameters

Generation Parameters:
rudetox - 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; \nsimplear - do_sample=false;until=["\n"]; \nchegeka - do_sample=false;until=["\n"];

The size of the context:
rudetox, ruhatespeech, rummlu, ruworldtree, ruopenbookqa, rumultiar, use, rwsd, mamuramu, multiq, rumodar, ruethics, ruhhh, rutie, ruhumaneval, rucodeeval, simplear, bps, lcs, chegeka, mathlogicqa, parus, rcb - 4096

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
Llama-2-13b-hf
MERA
0.067 0.133 0.367 0.1 0.067 0.233 0 - 0 0 0 0 0.033 0 0.1 0.067 0.033 0.033 0.033 0.033 0 0.033 0 0 0.033 0.108 0.1 0.067 0.033 0.1 0.067
Model, team Honest Helpful Harmless
Llama-2-13b-hf
MERA
0.459 0.508 0.5
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
Llama-2-13b-hf
MERA
0.415 0.404 0.434 0.585 0.471 0.647 0.505 0.524 0.488 0.502 0.246 0.278 0.29 0.509 0.496 0.428 0.4 0.382 0.244 0.466 0.242 0.62 0.31 0.376 0.179 0.44 0.327 0.491 0.514 0.26 0.58 0.636 0.423 0.606 0.457 0.321 0.32 0.442 0.291 0.379 0.53 0.338 0.469 0.329 0.504 0.278 0.505 0.281 0.298 0.398 0.36 0.578 0.379 0.378 0.47 0.527 0.513
Model, team SIM FL STA
Llama-2-13b-hf
MERA
0.88 0.574 0.111
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
Llama-2-13b-hf
MERA
0.356 0.505 0.433 0.435 0.421 0.5 0.448 0.526 0.538 0.585 0.462 0.442 0.45 0.488 0.544 0.346 0.533 0.444 0.421 0.614 0.281 0.61 0.333 0.521 0.333 0.485 0.5 0.526 0.754 0.533 0.644 0.487 0.366 0.491 0.424 0.536 0.244 0.689 0.404 0.585 0.555 0.46 0.533 0.533 0.793 0.622 0.672 0.409 0.585 0.702 0.689 0.609 0.608 0.468 0.419 0.474 0.578
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Llama-2-13b-hf
MERA
0.057 0.059 0.064 0.068 0.017
Model, team Virtue Law Moral Justice Utilitarianism
Llama-2-13b-hf
MERA
0.061 0.064 0.066 0.074 0.033
Model, team Virtue Law Moral Justice Utilitarianism
Llama-2-13b-hf
MERA
0.041 0.055 0.064 0.04 0.018
Model, team Women Men LGBT Nationalities Migrants Other
Llama-2-13b-hf
MERA
0.519 0.486 0.647 0.595 0.429 0.508