Yi-34B-Chat

MERA Created at 22.09.2024 22:09
0.425
The overall result
289
Place in the rating
Weak tasks:
310
RWSD
355
PARus
150
RCB
331
ruEthics
367
MultiQ
323
ruWorldTree
321
ruOpenBookQA
367
CheGeKa
336
ruMMLU
318
ruHateSpeech
309
ruDetox
246
ruHHH
233
ruTiE
99
ruHumanEval
290
USE
276
MathLogicQA
334
ruMultiAr
223
SimpleAr
402
LCS
423
BPS
161
ruModAr
327
MaMuRAMu
117
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.086 Accuracy
RCB 0.555 / 0.477 Accuracy F1 macro
USE 0.135 Grade norm
RWSD 0.523 Accuracy
PARus 0.688 Accuracy
ruTiE 0.687 Accuracy
MultiQ 0.295 / 0.113 F1 Exact match
CheGeKa 0.046 / 0.024 F1 Exact match
ruModAr 0.536 Exact match
MaMuRAMu 0.579 Accuracy
ruMultiAr 0.222 Exact match
ruCodeEval 0.174 / 0.313 / 0.366 Pass@k
MathLogicQA 0.409 Accuracy
ruWorldTree 0.783 / 0.78 Accuracy F1 macro
ruOpenBookQA 0.69 / 0.687 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 0.743 Accuracy
ruMMLU 0.491 Accuracy
SimpleAr 0.97 Exact match
ruHumanEval 0.2 / 0.308 / 0.354 Pass@k
ruHHH 0.685
ruHateSpeech 0.653
ruDetox 0.154
ruEthics
Correct God Ethical
Virtue 0.213 0.301 0.424
Law 0.262 0.354 0.449
Moral 0.266 0.364 0.503
Justice 0.201 0.282 0.414
Utilitarianism 0.163 0.26 0.341

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
bfloat16
Seed
1234
Batch
1
Transformers version
4.43.3
The number of GPUs and their type
4 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

MERA

Name of the ML model:

Yi-34B-Chat

Model size

34.4B

Model type:

Opened

SFT

Architecture description:

The Yi series models adopt the same model architecture as Llama but are NOT derivatives of Llama.

Description of the training:

Yi has independently created its own high-quality training datasets, efficient training pipelines, and robust training infrastructure entirely from the ground up.

Pretrain data:

Targeted as a bilingual language model and trained on 3T multilingual corpus with 4K sequence length and can be extended to 32K during inference time. Training data date up to June 2023. The released chat model has undergone exclusive training using Supervised Fine-Tuning (SFT).

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 - 4096

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
Yi-34B-Chat
MERA
0.3 0.1 0.633 0.133 0 0.267 0 - 0 0.033 0.033 0.033 0.1 0 0.167 0.25 0 0 0.033 0 0 0.167 0 0 0.067 0.158 0.2 0.1 0.333 0.433 0.333
Model, team Honest Helpful Harmless
Yi-34B-Chat
MERA
0.721 0.678 0.655
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
Yi-34B-Chat
MERA
0.393 0.458 0.586 0.735 0.516 0.682 0.68 0.595 0.485 0.534 0.404 0.413 0.35 0.611 0.584 0.581 0.57 0.458 0.233 0.519 0.334 0.532 0.26 0.387 0.357 0.48 0.37 0.602 0.661 0.23 0.68 0.752 0.472 0.687 0.449 0.479 0.29 0.603 0.377 0.433 0.662 0.368 0.497 0.496 0.577 0.412 0.647 0.293 0.401 0.461 0.45 0.696 0.495 0.521 0.73 0.679 0.56
Model, team SIM FL STA
Yi-34B-Chat
MERA
0.472 0.583 0.674
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
Yi-34B-Chat
MERA
0.378 0.535 0.4 0.602 0.487 0.69 0.431 0.544 0.462 0.615 0.667 0.6 0.458 0.651 0.503 0.481 0.57 0.6 0.368 0.667 0.404 0.746 0.578 0.533 0.644 0.5 0.628 0.579 0.825 0.6 0.756 0.718 0.563 0.772 0.485 0.554 0.467 0.711 0.561 0.446 0.624 0.476 0.689 0.667 0.845 0.711 0.603 0.568 0.554 0.754 0.867 0.536 0.646 0.623 0.512 0.433 0.644
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Yi-34B-Chat
MERA
0.213 0.262 0.266 0.201 0.163
Model, team Virtue Law Moral Justice Utilitarianism
Yi-34B-Chat
MERA
0.301 0.354 0.364 0.282 0.26
Model, team Virtue Law Moral Justice Utilitarianism
Yi-34B-Chat
MERA
0.424 0.449 0.503 0.414 0.341
Model, team Women Men LGBT Nationalities Migrants Other
Yi-34B-Chat
MERA
0.62 0.714 0.588 0.649 0.429 0.721