Qwen2.5-Coder-1.5B-Instruct

MERA Created at 04.11.2024 17:54
0.308
The overall result
470
Place in the rating
Weak tasks:
331
RWSD
534
PARus
448
RCB
513
MultiQ
500
ruWorldTree
508
ruOpenBookQA
535
CheGeKa
508
ruMMLU
468
ruHateSpeech
582
ruDetox
464
ruHHH
424
ruTiE
239
ruHumanEval
526
USE
490
MathLogicQA
353
ruMultiAr
335
SimpleAr
448
LCS
427
BPS
330
ruModAr
492
MaMuRAMu
236
ruCodeEval
+18
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.078 Accuracy
RCB 0.441 / 0.386 Accuracy F1 macro
USE 0.048 Grade norm
RWSD 0.519 Accuracy
PARus 0.5 Accuracy
ruTiE 0.538 Accuracy
MultiQ 0.18 / 0.112 F1 Exact match
CheGeKa 0.009 / 0 F1 Exact match
ruModAr 0.428 Exact match
MaMuRAMu 0.421 Accuracy
ruMultiAr 0.201 Exact match
ruCodeEval 0.007 / 0.035 / 0.067 Pass@k
MathLogicQA 0.303 Accuracy
ruWorldTree 0.537 / 0.529 Accuracy F1 macro
ruOpenBookQA 0.465 / 0.451 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 0.737 Accuracy
ruMMLU 0.35 Accuracy
SimpleAr 0.919 Exact match
ruHumanEval 0.02 / 0.087 / 0.152 Pass@k
ruHHH 0.506
ruHateSpeech 0.536
ruDetox 0.051
ruEthics
Correct God Ethical
Virtue 0.044 -0.012 0.058
Law 0.076 0.012 0.059
Moral 0.052 0.012 0.051
Justice 0.053 -0.028 0.052
Utilitarianism -0 -0.006 0.064

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-Coder-1.5B-Instruct

Model size

1.5B

Model type:

Opened

SFT

Additional links:

https://arxiv.org/pdf/2409.12186

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: 1.54B Number of Paramaters (Non-Embedding): 1.31B Number of Layers: 28 Number of Attention Heads (GQA): 12 for Q and 2 for KV Context Length: Full 131,072 tokens

Description of the training:

-

Pretrain data:

Large-scale, high-quality, and diverse data forms the foundation of pre-trained models. To this end, we constructed a dataset named Qwen2.5-Coder-Data. This dataset comprises five key data types: Source Code Data, Text-Code Grounding Data, Synthetic Data, Math Data, and Text Data, totaling 5.5 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 - 32768 \nrutie - 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-Coder-1.5B-Instruct
MERA
0.067 0.033 0.367 0.167 0 0.033 0 - 0.033 0 0 0 0 0 0.133 0.15 0.033 0 0.033 0.033 0 0 0 0 0.033 0.033 0.033 0.033 0.033 0.067 0.067
Model, team Honest Helpful Harmless
Qwen2.5-Coder-1.5B-Instruct
MERA
0.525 0.559 0.431
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-Coder-1.5B-Instruct
MERA
0.341 0.373 0.296 0.585 0.405 0.378 0.388 0.363 0.358 0.453 0.263 0.294 0.36 0.454 0.41 0.367 0.35 0.306 0.189 0.336 0.242 0.339 0.26 0.318 0.348 0.4 0.312 0.454 0.384 0.27 0.45 0.512 0.35 0.576 0.419 0.368 0.24 0.365 0.285 0.35 0.359 0.221 0.393 0.353 0.371 0.31 0.358 0.296 0.301 0.326 0.31 0.426 0.333 0.378 0.5 0.436 0.316
Model, team SIM FL STA
Qwen2.5-Coder-1.5B-Instruct
MERA
0.836 0.571 0.132
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-Coder-1.5B-Instruct
MERA
0.333 0.446 0.367 0.426 0.329 0.483 0.379 0.351 0.365 0.338 0.538 0.467 0.275 0.457 0.298 0.296 0.533 0.4 0.351 0.491 0.263 0.288 0.556 0.402 0.578 0.348 0.423 0.421 0.614 0.422 0.667 0.449 0.411 0.316 0.318 0.429 0.222 0.578 0.421 0.369 0.376 0.365 0.622 0.556 0.69 0.756 0.328 0.477 0.523 0.509 0.622 0.362 0.608 0.455 0.372 0.281 0.522
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-Coder-1.5B-Instruct
MERA
0.044 0.076 0.052 0.053 -0
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-Coder-1.5B-Instruct
MERA
-0.012 0.012 0.012 -0.028 -0.006
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2.5-Coder-1.5B-Instruct
MERA
0.058 0.059 0.051 0.052 0.064
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2.5-Coder-1.5B-Instruct
MERA
0.565 0.543 0.588 0.649 0.143 0.443