Qwen2.5-Math-1.5B

MERA Created at 04.11.2024 17:59
0.31
The overall result
468
Place in the rating
Weak tasks:
505
RWSD
555
PARus
521
RCB
585
MultiQ
562
ruWorldTree
532
ruOpenBookQA
575
CheGeKa
525
ruMMLU
572
ruHateSpeech
538
ruDetox
385
ruHHH
488
ruTiE
139
ruHumanEval
476
USE
358
MathLogicQA
125
ruMultiAr
227
SimpleAr
262
LCS
441
BPS
206
ruModAr
518
MaMuRAMu
164
ruCodeEval
+18
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.112 Accuracy
RCB 0.37 / 0.337 Accuracy F1 macro
USE 0.062 Grade norm
RWSD 0.473 Accuracy
PARus 0.496 Accuracy
ruTiE 0.508 Accuracy
MultiQ 0.095 / 0.06 F1 Exact match
CheGeKa 0.006 / 0 F1 Exact match
ruModAr 0.505 Exact match
MaMuRAMu 0.39 Accuracy
ruMultiAr 0.337 Exact match
ruCodeEval 0.08 / 0.21 / 0.287 Pass@k
MathLogicQA 0.361 Accuracy
ruWorldTree 0.377 / 0.365 Accuracy F1 macro
ruOpenBookQA 0.408 / 0.398 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 0.693 Accuracy
ruMMLU 0.329 Accuracy
SimpleAr 0.969 Exact match
ruHumanEval 0.141 / 0.301 / 0.378 Pass@k
ruHHH 0.556
ruHateSpeech 0.468
ruDetox 0.059
ruEthics
Correct God Ethical
Virtue 0.043 0.058 0.043
Law 0.055 0.052 0.019
Moral 0.037 0.05 0.037
Justice -0.029 0.037 0.023
Utilitarianism -0.006 0.025 0.079

Information about the submission

Mera version
v.1.2.0
Torch Version
2.4.0
The version of the codebase
9b26db97
CUDA version
12.1
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.44.2
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
vllm

Team:

MERA

Name of the ML model:

Qwen2.5-Math-1.5B

Model size

1.5B

Model type:

Opened

Pretrain

Additional links:

https://qwenlm.github.io/blog/qwen2.5-math/

Architecture description:

-

Description of the training:

-

Pretrain data:

First, the Qwen2-Math base models are trained on a high-quality mathematical pre-training dataset called the Qwen Math Corpus v1, which contains approximately 700 billion tokens. Second, we train a math-specific reward model Qwen2-Math-RM, derived from Qwen2-Math-72B, to create the Qwen2-Math-Instruct models. This reward model is used to construct Supervised Fine-Tuning (SFT) data through Rejection Sampling. Third, leveraging the Qwen2-Math-72B-Instruct model, we synthesize additional high-quality mathematical pre-training data, which serves as the foundation for Qwen Math Corpus v2. This updated corpus contains over 1 trillion tokens and is used to pre-train the Qwen2.5-Math models. Lastly, similar to the process used for the Qwen2-Math-Instruct models, we construct the Qwen2.5-Math-RM and Qwen2.5-Math-Instruct models.

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, ruhumaneval, rucodeeval, rutie - 4096

System prompt:
Реши задачу по инструкции ниже. Не давай никаких объяснений и пояснений к своему ответу. Не пиши ничего лишнего. Пиши только то, что указано в инструкции. Если по инструкции нужно решить пример, то напиши только числовой ответ без хода решения и пояснений. Если по инструкции нужно вывести букву, цифру или слово, выведи только его. Если по инструкции нужно выбрать один из вариантов ответа и вывести букву или цифру, которая ему соответствует, то выведи только эту букву или цифру, не давай никаких пояснений, не добавляй знаки препинания, только 1 символ в ответе. Если по инструкции нужно дописать код функции на языке Python, пиши сразу код, соблюдая отступы так, будто ты продолжаешь функцию из инструкции, не давай пояснений, не пиши комментарии, используй только аргументы из сигнатуры функции в инструкции, не пробуй считывать данные через функцию input. Не извиняйся, не строй диалог. Выдавай только ответ и ничего больше.

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
Qwen2.5-Math-1.5B
MERA
0 0 0.367 0.167 0 0.067 0 - 0 0 0 0 0 0 0.1 0.217 0 0 0.067 0 0 0.033 0 0 0.033 0.108 0.033 0.1 0.067 0 0.2
Model, team Honest Helpful Harmless
Qwen2.5-Math-1.5B
MERA
0.541 0.559 0.569
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
Qwen2.5-Math-1.5B
MERA
0.237 0.325 0.336 0.474 0.363 0.453 0.359 0.373 0.34 0.278 0.289 0.341 0.32 0.38 0.333 0.332 0.4 0.264 0.356 0.313 0.235 0.304 0.33 0.341 0.384 0.34 0.286 0.38 0.286 0.29 0.44 0.471 0.393 0.364 0.328 0.325 0.33 0.297 0.265 0.335 0.328 0.206 0.4 0.493 0.275 0.31 0.328 0.4 0.28 0.343 0.4 0.35 0.356 0.366 0.51 0.358 0.29
Model, team SIM FL STA
Qwen2.5-Math-1.5B
MERA
0.511 0.527 0.384
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
Qwen2.5-Math-1.5B
MERA
0.289 0.366 0.367 0.398 0.263 0.328 0.431 0.316 0.385 0.292 0.615 0.392 0.283 0.326 0.327 0.333 0.449 0.378 0.333 0.474 0.316 0.288 0.556 0.325 0.444 0.303 0.385 0.298 0.526 0.378 0.6 0.269 0.429 0.281 0.273 0.643 0.6 0.378 0.421 0.338 0.302 0.365 0.4 0.711 0.638 0.8 0.345 0.727 0.369 0.456 0.467 0.391 0.506 0.416 0.419 0.322 0.411
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-Math-1.5B
MERA
0.043 0.055 0.037 -0.029 -0.006
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-Math-1.5B
MERA
0.058 0.052 0.05 0.037 0.025
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-Math-1.5B
MERA
0.043 0.019 0.037 0.023 0.079
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2.5-Math-1.5B
MERA
0.454 0.514 0.412 0.432 0.714 0.475