Qwen2-1.5B

MERA Created at 22.09.2024 21:51
0.328
The overall result
441
Place in the rating
Weak tasks:
298
RWSD
572
PARus
591
RCB
496
ruEthics
468
MultiQ
392
ruWorldTree
402
ruOpenBookQA
410
CheGeKa
407
ruMMLU
423
ruHateSpeech
522
ruDetox
483
ruHHH
468
ruTiE
222
ruHumanEval
565
USE
373
MathLogicQA
399
ruMultiAr
329
SimpleAr
395
LCS
406
BPS
366
ruModAr
394
MaMuRAMu
246
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.088 Accuracy
RCB 0.331 / 0.282 Accuracy F1 macro
USE 0.037 Grade norm
RWSD 0.527 Accuracy
PARus 0.486 Accuracy
ruTiE 0.512 Accuracy
MultiQ 0.223 / 0.092 F1 Exact match
CheGeKa 0.03 / 0.01 F1 Exact match
ruModAr 0.396 Exact match
MaMuRAMu 0.51 Accuracy
ruMultiAr 0.169 Exact match
ruCodeEval 0.004 / 0.02 / 0.037 Pass@k
MathLogicQA 0.356 Accuracy
ruWorldTree 0.718 / 0.716 Accuracy F1 macro
ruOpenBookQA 0.625 / 0.626 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 0.773 Accuracy
ruMMLU 0.43 Accuracy
SimpleAr 0.923 Exact match
ruHumanEval 0.038 / 0.137 / 0.213 Pass@k
ruHHH 0.494
ruHateSpeech 0.558
ruDetox 0.066
ruEthics
Correct God Ethical
Virtue 0.03 0.079 0.069
Law 0.062 0.076 0.083
Moral 0.068 0.103 0.081
Justice 0.024 0.064 0.047
Utilitarianism 0.03 0.039 0.068

Information about the submission

Mera version
v.1.2.0
Torch Version
2.4.0
The version of the codebase
6ec745e0
CUDA version
12.1
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.43.2
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Qwen2-1.5B

Model size

1.5B

Model type:

Opened

Pretrain

Additional links:

https://arxiv.org/pdf/2407.10671

Architecture description:

Qwen2-1.5B is based on the Transformer architecture with SwiGLU activation, attention QKV bias, group query attention, RoPE and RMSNorm

Description of the training:

The model is trained using next-token prediction.

Pretrain data:

Qwen2-1.5B was pre-trained using the 7 trillion token dataset. The dataset supports approximately 30 languages, such as English, Chinese, Spanish, French, German, Arabic, Russian, Korean, Japanese, Thai, and Vietnamese. Similar to previous Qwen models, high-quality multi-task instruction data is integrated into the Qwen2 pre-training process to enhance in-context learning and instruction-following abilities.

License:

Apache-2.0

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
Qwen2-1.5B
MERA
0 0.067 0.3 0.133 0 0.067 0 - 0 0 0 0 0 0 0.133 0.133 0.033 0 0.033 0.033 0 0 0 0 0 0.008 0.067 0.033 0 0.033 0.033
Model, team Honest Helpful Harmless
Qwen2-1.5B
MERA
0.525 0.508 0.448
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
Qwen2-1.5B
MERA
0.415 0.373 0.441 0.667 0.529 0.647 0.621 0.505 0.438 0.43 0.263 0.31 0.29 0.519 0.496 0.488 0.47 0.382 0.289 0.458 0.274 0.526 0.32 0.428 0.295 0.45 0.351 0.426 0.486 0.36 0.61 0.686 0.448 0.616 0.502 0.385 0.29 0.468 0.338 0.414 0.591 0.353 0.49 0.361 0.521 0.375 0.471 0.319 0.34 0.395 0.41 0.532 0.444 0.462 0.49 0.539 0.44
Model, team SIM FL STA
Qwen2-1.5B
MERA
0.808 0.575 0.209
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
Qwen2-1.5B
MERA
0.422 0.525 0.417 0.519 0.368 0.569 0.466 0.386 0.385 0.462 0.603 0.5 0.408 0.62 0.368 0.506 0.57 0.467 0.474 0.544 0.316 0.542 0.333 0.467 0.556 0.591 0.615 0.491 0.754 0.533 0.733 0.679 0.518 0.456 0.409 0.554 0.4 0.578 0.316 0.508 0.527 0.524 0.6 0.511 0.828 0.756 0.5 0.318 0.677 0.702 0.6 0.333 0.646 0.558 0.349 0.398 0.578
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-1.5B
MERA
0.03 0.062 0.068 0.024 0.03
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-1.5B
MERA
0.079 0.076 0.103 0.064 0.039
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-1.5B
MERA
0.069 0.083 0.081 0.047 0.068
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2-1.5B
MERA
0.528 0.571 0.529 0.595 0.429 0.607