Qwen2.5-3B-Instruct

MERA Created at 04.11.2024 17:48
0.442
The overall result
255
Place in the rating
Weak tasks:
89
RWSD
297
PARus
267
RCB
369
ruEthics
250
MultiQ
277
ruWorldTree
336
ruOpenBookQA
379
CheGeKa
311
ruMMLU
225
ruHateSpeech
275
ruDetox
242
ruHHH
236
ruTiE
179
ruHumanEval
297
USE
255
MathLogicQA
200
ruMultiAr
266
SimpleAr
119
LCS
416
BPS
252
ruModAr
289
MaMuRAMu
204
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.148 Accuracy
RCB 0.527 / 0.484 Accuracy F1 macro
USE 0.132 Grade norm
RWSD 0.612 Accuracy
PARus 0.758 Accuracy
ruTiE 0.686 Accuracy
MultiQ 0.371 / 0.242 F1 Exact match
CheGeKa 0.04 / 0.026 F1 Exact match
ruModAr 0.479 Exact match
MaMuRAMu 0.621 Accuracy
ruMultiAr 0.289 Exact match
ruCodeEval 0.028 / 0.124 / 0.22 Pass@k
MathLogicQA 0.416 Accuracy
ruWorldTree 0.844 / 0.844 Accuracy F1 macro
ruOpenBookQA 0.683 / 0.683 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 0.752 Accuracy
ruMMLU 0.514 Accuracy
SimpleAr 0.948 Exact match
ruHumanEval 0.087 / 0.286 / 0.402 Pass@k
ruHHH 0.691
ruHateSpeech 0.743
ruDetox 0.172
ruEthics
Correct God Ethical
Virtue 0.169 0.191 0.162
Law 0.16 0.184 0.161
Moral 0.18 0.201 0.169
Justice 0.139 0.163 0.141
Utilitarianism 0.129 0.129 0.139

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-Instruct

Model size

3.0B

Model type:

Opened

SFT

Additional links:

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

Architecture description:

Type: Causal Language Models Training Stage: Pretraining & Post-training 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 and generation 8192 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, ruopenbookqa, rumultiar, use, rwsd, mamuramu, multiq, rumodar, ruethics, ruhhh, ruhumaneval - 32768 \nrutie - 15000 \nrutie, ruworldtree, rucodeeval - 10000

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-Instruct
MERA
0.167 0.133 0.633 0.2 0 0.1 0 - 0.033 0 0.033 0.067 0 0 0.1 0.35 0.033 0 0.133 0 0 0.233 0.067 0.1 0.067 0.217 0.067 0.1 0.2 0.167 0.3
Model, team Honest Helpful Harmless
Qwen2.5-3B-Instruct
MERA
0.672 0.661 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
Qwen2.5-3B-Instruct
MERA
0.481 0.41 0.625 0.739 0.601 0.642 0.67 0.601 0.531 0.538 0.342 0.381 0.29 0.602 0.616 0.575 0.59 0.479 0.356 0.595 0.245 0.678 0.33 0.503 0.384 0.56 0.362 0.611 0.624 0.42 0.66 0.645 0.521 0.747 0.525 0.479 0.37 0.665 0.444 0.483 0.712 0.496 0.593 0.496 0.691 0.491 0.662 0.378 0.387 0.489 0.5 0.713 0.551 0.525 0.65 0.709 0.534
Model, team SIM FL STA
Qwen2.5-3B-Instruct
MERA
0.605 0.635 0.533
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-Instruct
MERA
0.378 0.683 0.6 0.537 0.658 0.621 0.517 0.649 0.654 0.6 0.654 0.608 0.442 0.636 0.573 0.568 0.551 0.689 0.491 0.667 0.404 0.678 0.622 0.609 0.667 0.606 0.667 0.684 0.825 0.578 0.756 0.744 0.563 0.632 0.576 0.696 0.711 0.733 0.491 0.508 0.645 0.683 0.778 0.756 0.897 0.844 0.69 0.727 0.692 0.772 0.733 0.493 0.747 0.584 0.465 0.45 0.7
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-3B-Instruct
MERA
0.169 0.16 0.18 0.139 0.129
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-3B-Instruct
MERA
0.191 0.184 0.201 0.163 0.129
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-3B-Instruct
MERA
0.162 0.161 0.169 0.141 0.139
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2.5-3B-Instruct
MERA
0.787 0.771 0.588 0.757 0.429 0.721