Qwen2-Math-1.5B-Instruct

MERA Created at 22.09.2024 22:05
0.186
The overall result
610
Place in the rating
Weak tasks:
420
RWSD
554
PARus
590
RCB
628
MultiQ
599
ruWorldTree
606
ruOpenBookQA
582
ruMMLU
457
ruHateSpeech
444
ruDetox
541
ruHHH
533
ruTiE
521
USE
580
MathLogicQA
322
LCS
549
BPS
616
MaMuRAMu
+12
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.1 Accuracy
RCB 0.333 / 0.167 Accuracy F1 macro
USE 0.051 Grade norm
RWSD 0.5 Accuracy
PARus 0.498 Accuracy
ruTiE 0.504 Accuracy
MultiQ 0.004 / 0 F1 Exact match
CheGeKa 0 / 0 F1 Exact match
ruModAr 0 Exact match
MaMuRAMu 0.224 Accuracy
ruMultiAr 0 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.25 Accuracy
ruWorldTree 0.263 / 0.15 Accuracy F1 macro
ruOpenBookQA 0.26 / 0.153 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 0.509 Accuracy
ruMMLU 0.268 Accuracy
SimpleAr 0 Exact match
ruHumanEval 0 / 0 / 0 Pass@k
ruHHH 0.478
ruHateSpeech 0.543
ruDetox 0.091
ruEthics
Correct God Ethical
Virtue 0 0 0
Law 0 0 0
Moral 0 0 0
Justice 0 0 0
Utilitarianism 0 0 0

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
4.43.2
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
vllm

Team:

MERA

Name of the ML model:

Qwen2-Math-1.5B-Instruct

Model size

1.5B

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-1.5B-Instruct
MERA
0 0 0.367 0 0 0 0 - 0 0.067 0 0.033 0 0 0 0.267 0.033 0 0.033 0 0 0 0 0 0 0 0.167 0.033 0.033 0.2 0.233
Model, team Honest Helpful Harmless
Qwen2-Math-1.5B-Instruct
MERA
0.492 0.475 0.466
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-1.5B-Instruct
MERA
0.23 0.193 0.329 0.197 0.294 0.244 0.369 0.251 0.225 0.108 0.237 0.365 0.18 0.213 0.203 0.214 0.21 0.264 0.4 0.29 0.273 0.175 0.22 0.324 0.161 0.22 0.251 0.222 0.318 0.42 0.18 0.165 0.233 0.253 0.298 0.205 0.32 0.323 0.331 0.281 0.359 0.449 0.241 0.271 0.339 0.472 0.245 0.263 0.245 0.213 0.33 0.224 0.359 0.349 0.2 0.248 0.363
Model, team SIM FL STA
Qwen2-Math-1.5B-Instruct
MERA
0.219 0.735 0.543
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-1.5B-Instruct
MERA
0.178 0.218 0.183 0.259 0.263 0.172 0.259 0.228 0.269 0.215 0.218 0.258 0.158 0.194 0.24 0.247 0.28 0.311 0.175 0.175 0.228 0.288 0.244 0.225 0.222 0.273 0.244 0.228 0.175 0.156 0.222 0.205 0.143 0.281 0.258 0.214 0.267 0.356 0.14 0.154 0.241 0.238 0.133 0.222 0.241 0.111 0.259 0.205 0.277 0.175 0.267 0.217 0.253 0.26 0.233 0.199 0.2
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-Math-1.5B-Instruct
MERA
0 0 0 0 0
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-Math-1.5B-Instruct
MERA
0 0 0 0 0
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-Math-1.5B-Instruct
MERA
0 0 0 0 0
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2-Math-1.5B-Instruct
MERA
0.519 0.686 0.588 0.595 0.286 0.492