RU EN
Return to the leaderboard

Qwen2.5-Math-1.5B-Instruct

MERA Created at 04.11.2024 17:57
0.207
The overall result
517
Place in the rating
Weak tasks:
552
RWSD
476
PARus
554
RCB
551
MultiQ
551
ruWorldTree
516
ruOpenBookQA
557
CheGeKa
520
ruMMLU
532
ruHateSpeech
446
ruDetox
556
ruHHH
447
ruTiE
404
USE
334
MathLogicQA
238
ruMultiAr
213
SimpleAr
191
LCS
509
BPS
524
MaMuRAMu
+15
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.116 Accuracy
RCB 0.292 / 0.283 Accuracy F1 macro
USE 0.065 Grade norm
RWSD 0.235 Accuracy
PARus 0.5 Accuracy
ruTiE 0.505 Accuracy
MultiQ 0.008 / 0 F1 Exact match
CheGeKa 0.001 / 0 F1 Exact match
ruModAr 0 Exact match
MaMuRAMu 0.28 Accuracy
ruMultiAr 0.251 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.349 Accuracy
ruWorldTree 0.236 / 0.227 Accuracy F1 macro
ruOpenBookQA 0.293 / 0.283 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 0.482 Accuracy
ruMMLU 0.266 Accuracy
SimpleAr 0.953 Exact match
ruHumanEval 0 / 0 / 0 Pass@k
ruHHH 0.073
ruHateSpeech 0.442
ruDetox 0.069
ruEthics
Correct God Ethical
Virtue -0.022 0.002 0.007
Law -0.04 -0.02 -0.014
Moral -0.032 -0.004 -0.01
Justice -0.046 -0.006 -0.007
Utilitarianism -0.027 0.017 0.002

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
Butch
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-Instruct

Link to the ML model:

https://huggingface.co/Qwen/Qwen2.5-Math-1.5B-Instruct

Model size

1.5B

Model type:

Opened

SFT

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.

Expand information

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-Instruct
MERA 		0 0 0.2 0 0 0 0 - 0.067 0.033 0 0.067 0 0 0.1 0.383 0 0 0.033 0 0.067 0 0.033 0 0 0.092 0.1 0.133 0.1 0.067 0.067
Model, team Honest Helpful Harmless
Qwen2.5-Math-1.5B-Instruct
MERA 		0.066 0.085 0.069
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-Instruct
MERA 		0.304 0.283 0.257 0.265 0.301 0.299 0.252 0.299 0.275 0.26 0.193 0.333 0.34 0.269 0.272 0.295 0.24 0.264 0.278 0.29 0.241 0.222 0.31 0.243 0.152 0.24 0.256 0.222 0.245 0.17 0.3 0.388 0.294 0.333 0.26 0.244 0.22 0.319 0.219 0.32 0.278 0.346 0.248 0.231 0.287 0.259 0.216 0.196 0.206 0.28 0.27 0.186 0.3 0.303 0.33 0.242 0.269
Model, team SIM FL STA
Qwen2.5-Math-1.5B-Instruct
MERA 		0.231 0.479 0.608
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-Instruct
MERA 		0.156 0.317 0.283 0.222 0.355 0.328 0.362 0.228 0.385 0.323 0.359 0.267 0.342 0.248 0.222 0.235 0.299 0.289 0.263 0.281 0.211 0.305 0.289 0.343 0.244 0.182 0.282 0.175 0.596 0.178 0.4 0.231 0.17 0.193 0.273 0.286 0.356 0.244 0.316 0.262 0.224 0.159 0.289 0.556 0.397 0.422 0.224 0.5 0.246 0.228 0.178 0.348 0.329 0.351 0.302 0.199 0.256
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-Math-1.5B-Instruct
MERA 		-0.022 -0.04 -0.032 -0.046 -0.027
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-Math-1.5B-Instruct
MERA 		0.002 -0.02 -0.004 -0.006 0.017
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-Math-1.5B-Instruct
MERA 		0.007 -0.014 -0.01 -0.007 0.002
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2.5-Math-1.5B-Instruct
MERA 		0.37 0.486 0.353 0.486 0.571 0.525

Пользовательское соглашение

We use cookies to store data. By continuing to use the site, you consent to work with these files