Qwen1.5-0.5B

MERA Created at 22.09.2024 21:19
0.217
The overall result
571
Place in the rating
Weak tasks:
494
RWSD
574
PARus
568
RCB
599
MultiQ
588
ruWorldTree
585
ruOpenBookQA
568
CheGeKa
604
ruMMLU
432
ruHateSpeech
620
ruDetox
509
ruHHH
510
ruTiE
373
ruHumanEval
470
USE
535
MathLogicQA
457
ruMultiAr
447
SimpleAr
309
LCS
595
BPS
500
ruModAr
566
MaMuRAMu
257
ruCodeEval
+18
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.102 Accuracy
RCB 0.342 / 0.246 Accuracy F1 macro
USE 0.064 Grade norm
RWSD 0.481 Accuracy
PARus 0.486 Accuracy
ruTiE 0.505 Accuracy
MultiQ 0.064 / 0.014 F1 Exact match
CheGeKa 0.007 / 0 F1 Exact match
ruModAr 0.091 Exact match
MaMuRAMu 0.302 Accuracy
ruMultiAr 0.122 Exact match
ruCodeEval 0.003 / 0.014 / 0.024 Pass@k
MathLogicQA 0.276 Accuracy
ruWorldTree 0.276 / 0.173 Accuracy F1 macro
ruOpenBookQA 0.288 / 0.216 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 0.433 Accuracy
ruMMLU 0.24 Accuracy
SimpleAr 0.682 Exact match
ruHumanEval 0.002 / 0.009 / 0.018 Pass@k
ruHHH 0.483
ruHateSpeech 0.555
ruDetox 0.038
ruEthics
Correct God Ethical
Virtue 0.041 -0.042 -0.027
Law 0.055 -0.007 0.001
Moral 0.032 -0.024 -0.022
Justice 0.026 -0.034 -0.009
Utilitarianism 0.029 -0.035 -0.015

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:

Qwen1.5-0.5B

Model size

464.0M

Model type:

Opened

Pretrain

Additional links:

https://qwenlm.github.io/blog/qwen1.5/

Architecture description:

Qwen1.5-0.5B is based on the Transformer architecture with SwiGLU activation, attention QKV bias, group query attention, mixture of sliding window attention and full attention.

Description of the training:

The model is trained using next-token prediction.

Pretrain data:

Dataset consists of around 3T tokens and includes public web documents, encyclopedia, books, codes, etc. Additionally, dataset is multilingual, with a significant portion of the data being in English and Chinese. 1. To increase the diversity of our data, the deduplication techniques were employed, including exact-match deduplication after normalization and fuzzy deduplication using MinHash and LSH algorithms. 2. To filter out low-quality data, a combination of rule-based and machine-learning-based methods were used. 3. Texts are manually sampled from various sources and reviewed to ensure their quality. 4. To further enhance the performance of the model, high-quality instruction data have been incorporated into the pretraining process.

License:

tongyi-qianwen-research

Inference parameters

Generation Parameters:
rudetox - 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; \nsimplear - do_sample=false;until=["\n"]; \nchegeka - do_sample=false;until=["\n"];

The size of the context:
bps, lcs, mathlogicqa, rudetox, ruhatespeech, rummlu, ruworldtree, ruopenbookqa, rumultiar, use, rwsd, mamuramu, multiq, rumodar, ruethics, ruhhh, rutie, ruhumaneval, rucodeeval, simplear, chegeka, parus, rcb - 32768

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
Qwen1.5-0.5B
MERA
0.1 0 0.167 0.133 0 0 0 - 0 0.033 0 0 0 0 0.1 0.25 0.033 0 0.033 0.033 0 0 0 0 0 0.083 0.167 0.033 0.067 0.167 0.267
Model, team Honest Helpful Harmless
Qwen1.5-0.5B
MERA
0.492 0.492 0.466
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
Qwen1.5-0.5B
MERA
0.193 0.277 0.23 0.291 0.222 0.254 0.194 0.209 0.225 0.314 0.202 0.325 0.18 0.287 0.239 0.26 0.32 0.25 0.211 0.275 0.256 0.31 0.21 0.202 0.313 0.3 0.251 0.222 0.2 0.21 0.31 0.289 0.245 0.374 0.226 0.265 0.21 0.21 0.199 0.167 0.172 0.184 0.234 0.202 0.227 0.204 0.255 0.215 0.248 0.249 0.3 0.266 0.213 0.223 0.31 0.224 0.212
Model, team SIM FL STA
Qwen1.5-0.5B
MERA
0.744 0.534 0.167
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
Qwen1.5-0.5B
MERA
0.244 0.257 0.283 0.306 0.237 0.328 0.224 0.333 0.269 0.292 0.295 0.358 0.275 0.395 0.199 0.593 0.262 0.311 0.211 0.316 0.263 0.373 0.289 0.225 0.267 0.288 0.346 0.439 0.316 0.333 0.156 0.474 0.518 0.316 0.242 0.268 0.133 0.289 0.175 0.323 0.261 0.238 0.289 0.244 0.362 0.289 0.259 0.273 0.323 0.351 0.311 0.304 0.304 0.234 0.279 0.374 0.3
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen1.5-0.5B
MERA
0.041 0.055 0.032 0.026 0.029
Model, team Virtue Law Moral Justice Utilitarianism
Qwen1.5-0.5B
MERA
-0.042 -0.007 -0.024 -0.034 -0.035
Model, team Virtue Law Moral Justice Utilitarianism
Qwen1.5-0.5B
MERA
-0.027 0.001 -0.022 -0.009 -0.015
Model, team Women Men LGBT Nationalities Migrants Other
Qwen1.5-0.5B
MERA
0.528 0.629 0.588 0.595 0.286 0.557