Qwen2.5-3B

MERA Created at 04.11.2024 17:48
0.411
The overall result
315
Place in the rating
Weak tasks:
471
RWSD
353
PARus
411
RCB
421
ruEthics
375
MultiQ
294
ruWorldTree
301
ruOpenBookQA
363
CheGeKa
321
ruMMLU
371
ruHateSpeech
348
ruDetox
387
ruHHH
312
ruTiE
220
ruHumanEval
373
USE
277
MathLogicQA
188
ruMultiAr
257
SimpleAr
269
LCS
392
BPS
229
ruModAr
282
MaMuRAMu
221
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.11 Accuracy
RCB 0.473 / 0.431 Accuracy F1 macro
USE 0.086 Grade norm
RWSD 0.485 Accuracy
PARus 0.69 Accuracy
ruTiE 0.625 Accuracy
MultiQ 0.279 / 0.182 F1 Exact match
CheGeKa 0.048 / 0.026 F1 Exact match
ruModAr 0.489 Exact match
MaMuRAMu 0.628 Accuracy
ruMultiAr 0.294 Exact match
ruCodeEval 0.02 / 0.087 / 0.152 Pass@k
MathLogicQA 0.409 Accuracy
ruWorldTree 0.83 / 0.828 Accuracy F1 macro
ruOpenBookQA 0.71 / 0.707 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 0.798 Accuracy
ruMMLU 0.505 Accuracy
SimpleAr 0.955 Exact match
ruHumanEval 0.039 / 0.137 / 0.207 Pass@k
ruHHH 0.551
ruHateSpeech 0.611
ruDetox 0.137
ruEthics
Correct God Ethical
Virtue 0.15 0.14 0.135
Law 0.153 0.152 0.161
Moral 0.177 0.152 0.171
Justice 0.141 0.122 0.126
Utilitarianism 0.126 0.089 0.136

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:

Qwen2.5-3B

Model size

3.0B

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, Attention QKV bias and tied word embeddings Number of Parameters: 3.09B Number of Paramaters (Non-Embedding): 2.77B Number of Layers: 36 Number of Attention Heads (GQA): 16 for Q and 2 for KV Context Length: Full 32,768 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:

qwen-research

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 - 32768 \nrutie - 15000

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
Qwen2.5-3B
MERA
0.3 0.1 0.5 0.167 0 0.033 0 - 0 0 0 0 0.1 0 0.133 0.083 0.033 0.033 0.067 0 0 0.3 0 0 0.033 0.2 0.033 0.033 0 0.033 0.067
Model, team Honest Helpful Harmless
Qwen2.5-3B
MERA
0.475 0.576 0.603
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
Qwen2.5-3B
MERA
0.452 0.434 0.592 0.726 0.588 0.642 0.612 0.556 0.562 0.592 0.316 0.333 0.33 0.62 0.594 0.566 0.56 0.444 0.356 0.55 0.296 0.661 0.34 0.451 0.375 0.56 0.379 0.528 0.559 0.41 0.67 0.711 0.485 0.727 0.528 0.491 0.39 0.629 0.377 0.488 0.652 0.415 0.545 0.491 0.633 0.412 0.657 0.422 0.383 0.507 0.44 0.671 0.515 0.504 0.68 0.685 0.518
Model, team SIM FL STA
Qwen2.5-3B
MERA
0.555 0.694 0.48
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
Qwen2.5-3B
MERA
0.444 0.693 0.6 0.574 0.684 0.638 0.5 0.667 0.673 0.631 0.718 0.567 0.442 0.667 0.567 0.519 0.579 0.667 0.491 0.754 0.439 0.729 0.778 0.633 0.667 0.545 0.667 0.632 0.807 0.689 0.733 0.667 0.554 0.719 0.485 0.679 0.667 0.756 0.579 0.585 0.629 0.619 0.867 0.822 0.914 0.8 0.741 0.773 0.631 0.772 0.756 0.522 0.709 0.584 0.558 0.444 0.656
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-3B
MERA
0.15 0.153 0.177 0.141 0.126
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-3B
MERA
0.14 0.152 0.152 0.122 0.089
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-3B
MERA
0.135 0.161 0.171 0.126 0.136
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2.5-3B
MERA
0.583 0.686 0.529 0.757 0.429 0.574