Qwen1.5-7B

MERA Created at 22.09.2024 21:40
0.374
The overall result
362
Place in the rating
Weak tasks:
206
RWSD
432
PARus
414
RCB
437
ruEthics
296
MultiQ
399
ruWorldTree
370
ruOpenBookQA
371
CheGeKa
392
ruMMLU
349
ruHateSpeech
412
ruDetox
351
ruHHH
393
ruTiE
325
ruHumanEval
502
USE
314
MathLogicQA
236
ruMultiAr
241
SimpleAr
472
LCS
410
BPS
233
ruModAr
382
MaMuRAMu
290
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.072 Accuracy
RCB 0.473 / 0.419 Accuracy F1 macro
USE 0.056 Grade norm
RWSD 0.558 Accuracy
PARus 0.6 Accuracy
ruTiE 0.556 Accuracy
MultiQ 0.345 / 0.17 F1 Exact match
CheGeKa 0.043 / 0.022 F1 Exact match
ruModAr 0.485 Exact match
MaMuRAMu 0.52 Accuracy
ruMultiAr 0.275 Exact match
ruCodeEval 0.001 / 0.003 / 0.006 Pass@k
MathLogicQA 0.382 Accuracy
ruWorldTree 0.714 / 0.711 Accuracy F1 macro
ruOpenBookQA 0.65 / 0.65 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 0.764 Accuracy
ruMMLU 0.441 Accuracy
SimpleAr 0.965 Exact match
ruHumanEval 0.005 / 0.026 / 0.049 Pass@k
ruHHH 0.59
ruHateSpeech 0.634
ruDetox 0.105
ruEthics
Correct God Ethical
Virtue 0.105 0.121 0.139
Law 0.108 0.137 0.168
Moral 0.115 0.134 0.173
Justice 0.079 0.107 0.128
Utilitarianism 0.077 0.1 0.144

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:

Qwen1.5-7B

Model size

7.7B

Model type:

Opened

Pretrain

Additional links:

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

Architecture description:

Qwen1.5-7B 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:
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, rutie, ruhumaneval, rucodeeval - 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
Qwen1.5-7B
MERA
0.1 0.1 0.333 0.067 0 0.1 0 - 0 0 0 0.033 0 0 0.133 0.067 0.033 0.033 0.067 0.033 0 0 0 0 0 0.1 0.067 0.067 0.033 0.067 0.1
Model, team Honest Helpful Harmless
Qwen1.5-7B
MERA
0.541 0.627 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
Qwen1.5-7B
MERA
0.378 0.404 0.48 0.692 0.49 0.622 0.553 0.514 0.481 0.439 0.342 0.349 0.38 0.509 0.497 0.532 0.52 0.431 0.278 0.58 0.253 0.497 0.26 0.405 0.313 0.47 0.364 0.509 0.555 0.4 0.58 0.645 0.448 0.657 0.528 0.427 0.37 0.484 0.351 0.438 0.551 0.338 0.49 0.408 0.509 0.361 0.51 0.274 0.358 0.4 0.42 0.603 0.415 0.424 0.59 0.527 0.497
Model, team SIM FL STA
Qwen1.5-7B
MERA
0.727 0.581 0.323
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
Qwen1.5-7B
MERA
0.311 0.554 0.417 0.509 0.421 0.569 0.483 0.544 0.385 0.492 0.59 0.483 0.3 0.589 0.427 0.432 0.561 0.444 0.386 0.561 0.316 0.61 0.444 0.503 0.644 0.576 0.526 0.474 0.789 0.622 0.733 0.628 0.527 0.649 0.409 0.464 0.422 0.622 0.404 0.492 0.559 0.476 0.778 0.556 0.81 0.6 0.569 0.545 0.615 0.684 0.6 0.42 0.633 0.649 0.372 0.38 0.556
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen1.5-7B
MERA
0.105 0.108 0.115 0.079 0.077
Model, team Virtue Law Moral Justice Utilitarianism
Qwen1.5-7B
MERA
0.121 0.137 0.134 0.107 0.1
Model, team Virtue Law Moral Justice Utilitarianism
Qwen1.5-7B
MERA
0.139 0.168 0.173 0.128 0.144
Model, team Women Men LGBT Nationalities Migrants Other
Qwen1.5-7B
MERA
0.611 0.714 0.647 0.622 0.571 0.639