Qwen2-0.5B

MERA Created at 22.09.2024 21:47
0.248
The overall result
540
Place in the rating
Weak tasks:
533
RWSD
546
PARus
593
RCB
569
MultiQ
534
ruWorldTree
558
ruOpenBookQA
541
CheGeKa
540
ruMMLU
414
ruHateSpeech
611
ruDetox
520
ruHHH
521
ruTiE
228
ruHumanEval
584
USE
465
MathLogicQA
451
ruMultiAr
449
SimpleAr
445
LCS
588
BPS
444
ruModAr
545
MaMuRAMu
243
ruCodeEval
+18
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.08 Accuracy
RCB 0.331 / 0.217 Accuracy F1 macro
USE 0.025 Grade norm
RWSD 0.458 Accuracy
PARus 0.498 Accuracy
ruTiE 0.504 Accuracy
MultiQ 0.119 / 0.039 F1 Exact match
CheGeKa 0.009 / 0 F1 Exact match
ruModAr 0.209 Exact match
MaMuRAMu 0.335 Accuracy
ruMultiAr 0.125 Exact match
ruCodeEval 0.005 / 0.02 / 0.03 Pass@k
MathLogicQA 0.312 Accuracy
ruWorldTree 0.44 / 0.423 Accuracy F1 macro
ruOpenBookQA 0.368 / 0.357 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 0.464 Accuracy
ruMMLU 0.319 Accuracy
SimpleAr 0.672 Exact match
ruHumanEval 0.036 / 0.112 / 0.165 Pass@k
ruHHH 0.478
ruHateSpeech 0.566
ruDetox 0.042
ruEthics
Correct God Ethical
Virtue -0.009 0.035 0.042
Law 0.023 0.035 0.054
Moral 0.023 0.042 0.043
Justice 0.038 0.031 0.057
Utilitarianism -0.006 0.052 0.032

Information about the submission

Mera version
v.1.2.0
Torch Version
2.4.0
The version of the codebase
6ec745e0
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
hf

Team:

MERA

Name of the ML model:

Qwen2-0.5B

Model size

494.0M

Model type:

Opened

Pretrain

Additional links:

https://arxiv.org/pdf/2407.10671

Architecture description:

Qwen2-0.5B is based on the Transformer architecture with SwiGLU activation, attention QKV bias, group query attention, RoPE and RMSNorm

Description of the training:

The model is trained using next-token prediction.

Pretrain data:

Qwen2-0.5B was pre-trained using the 12 trillion token dataset. The dataset supports approximately 30 languages, such as English, Chinese, Spanish, French, German, Arabic, Russian, Korean, Japanese, Thai, and Vietnamese. Similar to previous Qwen models, high-quality multi-task instruction data is integrated into the Qwen2 pre-training process to enhance in-context learning and instruction-following abilities.

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:
mathlogicqa, parus, rcb, simplear, bps, lcs, chegeka, mathlogicqa, parus, rcb, rudetox, ruhatespeech, rummlu, ruworldtree, ruopenbookqa, rumultiar, use, rwsd, mamuramu, multiq, rumodar, ruethics, ruhhh, rutie, ruhumaneval, rucodeeval - 131072

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
Qwen2-0.5B
MERA
0 0.033 0.233 0.133 0 0.133 0 - 0 0 0 0 0 0 0.033 0.033 0.033 0 0.067 0.033 0 0 0 0 0 0 0 0.033 0.033 0 0.033
Model, team Honest Helpful Harmless
Qwen2-0.5B
MERA
0.492 0.492 0.448
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
Qwen2-0.5B
MERA
0.37 0.211 0.289 0.432 0.379 0.393 0.388 0.338 0.392 0.309 0.175 0.254 0.36 0.454 0.345 0.361 0.32 0.278 0.244 0.313 0.28 0.228 0.23 0.335 0.286 0.35 0.297 0.361 0.241 0.3 0.34 0.512 0.319 0.475 0.321 0.278 0.29 0.358 0.245 0.291 0.323 0.294 0.352 0.276 0.318 0.338 0.304 0.252 0.326 0.311 0.31 0.304 0.328 0.303 0.39 0.412 0.373
Model, team SIM FL STA
Qwen2-0.5B
MERA
0.782 0.546 0.146
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
Qwen2-0.5B
MERA
0.289 0.307 0.283 0.389 0.276 0.328 0.414 0.246 0.288 0.354 0.346 0.283 0.233 0.349 0.234 0.42 0.393 0.378 0.368 0.404 0.228 0.186 0.356 0.331 0.444 0.409 0.449 0.228 0.579 0.244 0.6 0.41 0.429 0.281 0.303 0.339 0.2 0.333 0.386 0.415 0.253 0.317 0.378 0.378 0.534 0.511 0.259 0.318 0.385 0.386 0.333 0.304 0.468 0.312 0.186 0.246 0.356
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-0.5B
MERA
-0.009 0.023 0.023 0.038 -0.006
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-0.5B
MERA
0.035 0.035 0.042 0.031 0.052
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-0.5B
MERA
0.042 0.054 0.043 0.057 0.032
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2-0.5B
MERA
0.509 0.686 0.529 0.676 0.286 0.574