Qwen2-7B

MERA Created at 20.09.2024 19:40
0.455
The overall result
229
Place in the rating
Weak tasks:
153
RWSD
259
PARus
316
RCB
277
ruEthics
350
MultiQ
202
ruWorldTree
241
ruOpenBookQA
286
CheGeKa
237
ruMMLU
266
ruHateSpeech
257
ruDetox
258
ruHHH
203
ruTiE
150
ruHumanEval
256
USE
139
MathLogicQA
234
ruMultiAr
126
SimpleAr
439
LCS
273
BPS
214
ruModAr
246
MaMuRAMu
225
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.08 Accuracy
RCB 0.511 / 0.438 Accuracy F1 macro
USE 0.157 Grade norm
RWSD 0.577 Accuracy
PARus 0.802 Accuracy
ruTiE 0.709 Accuracy
MultiQ 0.305 / 0.14 F1 Exact match
CheGeKa 0.1 / 0.067 F1 Exact match
ruModAr 0.497 Exact match
MaMuRAMu 0.685 Accuracy
ruMultiAr 0.275 Exact match
ruCodeEval 0.017 / 0.073 / 0.116 Pass@k
MathLogicQA 0.481 Accuracy
ruWorldTree 0.928 / 0.928 Accuracy F1 macro
ruOpenBookQA 0.79 / 0.79 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 0.926 Accuracy
ruMMLU 0.581 Accuracy
SimpleAr 0.987 Exact match
ruHumanEval 0.114 / 0.325 / 0.439 Pass@k
ruHHH 0.669
ruHateSpeech 0.702
ruDetox 0.177
ruEthics
Correct God Ethical
Virtue 0.27 0.293 0.3
Law 0.273 0.304 0.309
Moral 0.297 0.325 0.336
Justice 0.225 0.263 0.276
Utilitarianism 0.208 0.259 0.258

Information about the submission

Mera version
v.1.2.0
Torch Version
2.3.1
The version of the codebase
aec92f8
CUDA version
12.1
Precision of the model weights
float16
Seed
1234
Batch
1
Transformers version
4.42.3
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Qwen2-7B

Link to the ML model:

https://huggingface.co/Qwen/Qwen2-7B

Model size

7.6B

Model type:

Opened

Pretrain

Additional links:

https://arxiv.org/pdf/2407.10671

Architecture description:

Qwen2-7B 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:

Qwen2-7B was pre-trained using the 7 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:
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
Qwen2-7B
MERA
0.267 0.2 0.667 0.2 0.033 0.133 0 - 0 0.1 0 0 0.133 0 0.1 0.233 0.033 0 0 0 0 0.567 0.133 0.067 0.033 0.375 0.033 0.133 0.267 0.033 0.233
Model, team Honest Helpful Harmless
Qwen2-7B
MERA
0.508 0.729 0.776
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
Qwen2-7B
MERA
0.556 0.494 0.658 0.825 0.654 0.761 0.699 0.666 0.614 0.619 0.456 0.397 0.44 0.667 0.708 0.653 0.72 0.639 0.367 0.618 0.251 0.702 0.43 0.538 0.402 0.66 0.424 0.611 0.718 0.38 0.73 0.736 0.571 0.818 0.608 0.628 0.41 0.758 0.364 0.488 0.753 0.585 0.579 0.546 0.745 0.532 0.73 0.448 0.433 0.587 0.54 0.743 0.669 0.664 0.71 0.764 0.767
Model, team SIM FL STA
Qwen2-7B
MERA
0.606 0.702 0.521
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
Qwen2-7B
MERA
0.467 0.772 0.6 0.657 0.737 0.724 0.534 0.596 0.596 0.662 0.641 0.65 0.417 0.736 0.661 0.642 0.682 0.733 0.509 0.807 0.316 0.831 0.778 0.728 0.644 0.697 0.718 0.614 0.912 0.622 0.844 0.782 0.625 0.825 0.652 0.661 0.689 0.778 0.596 0.692 0.763 0.698 0.756 0.756 0.879 0.822 0.897 0.795 0.692 0.842 0.689 0.623 0.873 0.571 0.535 0.52 0.778
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-7B
MERA
0.27 0.273 0.297 0.225 0.208
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-7B
MERA
0.293 0.304 0.325 0.263 0.259
Model, team Virtue Law Moral Justice Utilitarianism
Qwen2-7B
MERA
0.3 0.309 0.336 0.276 0.258
Model, team Women Men LGBT Nationalities Migrants Other
Qwen2-7B
MERA
0.704 0.6 0.765 0.73 0.429 0.754