Qwen2-Math-7B-Instruct

MERA Created at 22.09.2024 22:05
0.345
The overall result
409
Place in the rating
Weak tasks:
275
RWSD
419
PARus
406
RCB
378
ruEthics
517
MultiQ
445
ruWorldTree
463
ruOpenBookQA
573
CheGeKa
473
ruMMLU
358
ruHateSpeech
546
ruDetox
429
ruHHH
380
ruTiE
314
ruHumanEval
472
USE
301
MathLogicQA
316
ruMultiAr
366
SimpleAr
207
LCS
373
BPS
298
ruModAr
447
MaMuRAMu
+18
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.124 Accuracy
RCB 0.477 / 0.425 Accuracy F1 macro
USE 0.063 Grade norm
RWSD 0.535 Accuracy
PARus 0.616 Accuracy
ruTiE 0.562 Accuracy
MultiQ 0.177 / 0.092 F1 Exact match
CheGeKa 0.006 / 0 F1 Exact match
ruModAr 0.461 Exact match
MaMuRAMu 0.454 Accuracy
ruMultiAr 0.236 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.393 Accuracy
ruWorldTree 0.627 / 0.616 Accuracy F1 macro
ruOpenBookQA 0.53 / 0.526 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 0.835 Accuracy
ruMMLU 0.377 Accuracy
SimpleAr 0.892 Exact match
ruHumanEval 0.006 / 0.006 / 0.006 Pass@k
ruHHH 0.528
ruHateSpeech 0.626
ruDetox 0.056
ruEthics
Correct God Ethical
Virtue 0.18 0.293 0.263
Law 0.144 0.268 0.253
Moral 0.172 0.287 0.279
Justice 0.149 0.258 0.228
Utilitarianism 0.121 0.25 0.237

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
-
Seed
1234
Batch
1
Transformers version
4.43.2
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
vllm

Team:

MERA

Name of the ML model:

Qwen2-Math-7B-Instruct

Model type:

Opened

SFT

Additional links:

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

Architecture description:

This is a specialized math language models built upon the Qwen2 LLMs, which significantly outperforms the mathematical capabilities of open-source models and even closed-source models (e.g., GPT4o).

Description of the training:

The model was trained a math-specific reward model based on Qwen2-Math-72B. Then the authors combined this dense reward signal with a binary signal indicating whether the model answered correctly. This combined signal is used as supervision for constructing the SFT data through Rejection Sampling and also in the reinforcement learning with Group Relative Policy Optimization (GRPO) after SFT.

Pretrain data:

The base models of Qwen2-Math is initialized with Qwen2, and then pretrained on a meticulously designed Mathematics-specific Corpus. This corpus contains large-scale high-quality mathematical web texts, books, codes, exam questions, and mathematical pre-training data synthesized by Qwen2.

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-Math-7B-Instruct
MERA
0.2 0 0.2 0.067 0 0.167 0 - 0 0 0 0 0 0 0.133 0.2 0.033 0.033 0 0.033 0 0 0 0 0.033 0.142 0.133 0.067 0 0.033 0.033
Model, team Honest Helpful Harmless
Qwen2-Math-7B-Instruct
MERA
0.525 0.627 0.431
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-Math-7B-Instruct
MERA
0.333 0.343 0.388 0.56 0.379 0.438 0.505 0.408 0.346 0.457 0.404 0.381 0.32 0.463 0.462 0.376 0.4 0.319 0.267 0.397 0.246 0.269 0.37 0.283 0.384 0.42 0.269 0.463 0.367 0.33 0.51 0.397 0.399 0.465 0.366 0.449 0.37 0.439 0.391 0.433 0.424 0.283 0.4 0.491 0.468 0.37 0.333 0.374 0.348 0.356 0.38 0.418 0.382 0.462 0.57 0.479 0.311
Model, team SIM FL STA
Qwen2-Math-7B-Instruct
MERA
0.661 0.521 0.258
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-Math-7B-Instruct
MERA
0.289 0.475 0.417 0.472 0.368 0.362 0.414 0.421 0.327 0.431 0.654 0.592 0.225 0.45 0.275 0.383 0.533 0.489 0.456 0.474 0.281 0.305 0.6 0.402 0.6 0.348 0.397 0.404 0.596 0.467 0.667 0.474 0.42 0.368 0.364 0.554 0.644 0.467 0.526 0.477 0.4 0.476 0.6 0.733 0.655 0.889 0.448 0.795 0.6 0.526 0.622 0.319 0.671 0.468 0.302 0.31 0.467
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-Math-7B-Instruct
MERA
0.18 0.144 0.172 0.149 0.121
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-Math-7B-Instruct
MERA
0.293 0.268 0.287 0.258 0.25
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-Math-7B-Instruct
MERA
0.263 0.253 0.279 0.228 0.237
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2-Math-7B-Instruct
MERA
0.583 0.686 0.588 0.703 0.286 0.672