Qwen2.5-7B

MERA Created at 04.11.2024 17:52
0.456
The overall result
227
Place in the rating
Weak tasks:
308
RWSD
270
PARus
240
RCB
402
ruEthics
252
MultiQ
235
ruWorldTree
224
ruOpenBookQA
297
CheGeKa
218
ruMMLU
261
ruHateSpeech
195
ruDetox
248
ruHHH
347
ruTiE
199
ruHumanEval
260
USE
201
MathLogicQA
163
ruMultiAr
93
SimpleAr
519
LCS
152
BPS
159
ruModAr
199
MaMuRAMu
206
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.054 Accuracy
RCB 0.534 / 0.462 Accuracy F1 macro
USE 0.156 Grade norm
RWSD 0.523 Accuracy
PARus 0.794 Accuracy
ruTiE 0.589 Accuracy
MultiQ 0.37 / 0.247 F1 Exact match
CheGeKa 0.09 / 0.065 F1 Exact match
ruModAr 0.54 Exact match
MaMuRAMu 0.713 Accuracy
ruMultiAr 0.308 Exact match
ruCodeEval 0.026 / 0.109 / 0.183 Pass@k
MathLogicQA 0.455 Accuracy
ruWorldTree 0.903 / 0.902 Accuracy F1 macro
ruOpenBookQA 0.818 / 0.818 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 0.966 Accuracy
ruMMLU 0.596 Accuracy
SimpleAr 0.991 Exact match
ruHumanEval 0.062 / 0.225 / 0.354 Pass@k
ruHHH 0.68
ruHateSpeech 0.709
ruDetox 0.208
ruEthics
Correct God Ethical
Virtue 0.112 0.241 0.191
Law 0.146 0.246 0.176
Moral 0.152 0.255 0.185
Justice 0.106 0.212 0.175
Utilitarianism 0.103 0.212 0.177

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:

Qwen2.5-7B

Model size

7.0B

Model type:

Opened

Pretrain

Additional links:

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

Architecture description:

Type: Causal Language Models Training Stage: Pretraining Architecture: transformers with RoPE, SwiGLU, RMSNorm, and Attention QKV bias Number of Parameters: 7.61B Number of Paramaters (Non-Embedding): 6.53B Number of Layers: 28 Number of Attention Heads (GQA): 28 for Q and 4 for KV Context Length: 131,072 tokens

Description of the training:

-

Pretrain data:

The size of the pre-training dataset is expanded from 7 trillion tokens to a maximum of 18 trillion tokens.

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 - 131072

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
Qwen2.5-7B
MERA
0.267 0.4 0.767 0.167 0 0.233 0 - 0 0.033 0 0 0.067 0.033 0.1 0.117 0.033 0 0 0.033 0 0.6 0.1 0.033 0.033 0.317 0.133 0.1 0.367 0.033 0.267
Model, team Honest Helpful Harmless
Qwen2.5-7B
MERA
0.59 0.712 0.741
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
Qwen2.5-7B
MERA
0.615 0.512 0.743 0.795 0.673 0.741 0.709 0.659 0.648 0.668 0.535 0.397 0.45 0.639 0.702 0.688 0.7 0.632 0.467 0.656 0.239 0.772 0.43 0.572 0.473 0.67 0.426 0.574 0.682 0.45 0.76 0.752 0.571 0.788 0.672 0.645 0.46 0.768 0.503 0.586 0.778 0.57 0.676 0.626 0.73 0.537 0.74 0.452 0.411 0.587 0.59 0.797 0.672 0.702 0.8 0.715 0.705
Model, team SIM FL STA
Qwen2.5-7B
MERA
0.578 0.707 0.611
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
Qwen2.5-7B
MERA
0.667 0.802 0.6 0.639 0.763 0.69 0.569 0.684 0.692 0.708 0.718 0.667 0.533 0.721 0.637 0.679 0.682 0.733 0.544 0.807 0.404 0.797 0.822 0.757 0.756 0.697 0.744 0.632 0.895 0.778 0.8 0.769 0.616 0.86 0.636 0.75 0.822 0.733 0.596 0.662 0.812 0.778 0.8 0.8 0.845 0.889 0.862 0.795 0.662 0.825 0.867 0.725 0.81 0.662 0.605 0.538 0.811
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-7B
MERA
0.112 0.146 0.152 0.106 0.103
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-7B
MERA
0.241 0.246 0.255 0.212 0.212
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-7B
MERA
0.191 0.176 0.185 0.175 0.177
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2.5-7B
MERA
0.694 0.686 0.706 0.784 0.286 0.754