Llama-3.2-3B

MERA Created at 04.11.2024 17:39
0.362
The overall result
375
Place in the rating
Weak tasks:
495
RWSD
453
PARus
461
RCB
479
ruEthics
430
MultiQ
423
ruWorldTree
410
ruOpenBookQA
345
CheGeKa
411
ruMMLU
439
ruHateSpeech
572
ruDetox
511
ruHHH
498
ruTiE
141
ruHumanEval
411
USE
470
MathLogicQA
403
ruMultiAr
398
SimpleAr
293
LCS
451
BPS
146
ruModAr
377
MaMuRAMu
177
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.106 Accuracy
RCB 0.422 / 0.383 Accuracy F1 macro
USE 0.075 Grade norm
RWSD 0.477 Accuracy
PARus 0.564 Accuracy
ruTiE 0.506 Accuracy
MultiQ 0.248 / 0.132 F1 Exact match
CheGeKa 0.06 / 0.041 F1 Exact match
ruModAr 0.592 Exact match
MaMuRAMu 0.528 Accuracy
ruMultiAr 0.165 Exact match
ruCodeEval 0.063 / 0.192 / 0.274 Pass@k
MathLogicQA 0.309 Accuracy
ruWorldTree 0.67 / 0.669 Accuracy F1 macro
ruOpenBookQA 0.615 / 0.612 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 0.669 Accuracy
ruMMLU 0.427 Accuracy
SimpleAr 0.845 Exact match
ruHumanEval 0.14 / 0.286 / 0.348 Pass@k
ruHHH 0.478
ruHateSpeech 0.547
ruDetox 0.052
ruEthics
Correct God Ethical
Virtue 0.055 0.096 0.054
Law 0.067 0.113 0.068
Moral 0.077 0.111 0.096
Justice 0.041 0.094 0.047
Utilitarianism 0.056 0.084 0.049

Information about the submission

Mera version
v.1.2.0
Torch Version
2.4.0
The version of the codebase
9b26db97
CUDA version
12.1
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.44.2
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
vllm

Team:

MERA

Name of the ML model:

Llama-3.2-3B

Model size

3.0B

Model type:

Opened

Pretrain

Architecture description:

Llama 3.2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.

Description of the training:

-

Pretrain data:

Llama 3.2 was pretrained on up to 9 trillion tokens of data from publicly available sources. For the 1B and 3B Llama 3.2 models, we incorporated logits from the Llama 3.1 8B and 70B models into the pretraining stage of the model development, where outputs (logits) from these larger models were used as token-level targets. Knowledge distillation was used after pruning to recover performance. In post-training we used a similar recipe as Llama 3.1 and produced final chat models by doing several rounds of alignment on top of the pre-trained model. Each round involved Supervised Fine-Tuning (SFT), Rejection Sampling (RS), and Direct Preference Optimization (DPO).

License:

llama3.2

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, ruhumaneval, rucodeeval, rutie - 131072

System prompt:
Реши задачу по инструкции ниже. Не давай никаких объяснений и пояснений к своему ответу. Не пиши ничего лишнего. Пиши только то, что указано в инструкции. Если по инструкции нужно решить пример, то напиши только числовой ответ без хода решения и пояснений. Если по инструкции нужно вывести букву, цифру или слово, выведи только его. Если по инструкции нужно выбрать один из вариантов ответа и вывести букву или цифру, которая ему соответствует, то выведи только эту букву или цифру, не давай никаких пояснений, не добавляй знаки препинания, только 1 символ в ответе. Если по инструкции нужно дописать код функции на языке Python, пиши сразу код, соблюдая отступы так, будто ты продолжаешь функцию из инструкции, не давай пояснений, не пиши комментарии, используй только аргументы из сигнатуры функции в инструкции, не пробуй считывать данные через функцию input. Не извиняйся, не строй диалог. Выдавай только ответ и ничего больше.

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
Llama-3.2-3B
MERA
0.067 0.033 0.433 0.133 0.033 0.167 0 - 0 0 0 0 0.033 0 0.1 0.267 0.033 0.033 0.033 0 0 0.067 0 0 0.1 0.033 0.1 0.167 0.1 0.1 0.167
Model, team Honest Helpful Harmless
Llama-3.2-3B
MERA
0.41 0.542 0.483
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
Llama-3.2-3B
MERA
0.452 0.386 0.5 0.654 0.467 0.607 0.505 0.492 0.481 0.435 0.281 0.294 0.31 0.426 0.511 0.497 0.47 0.41 0.267 0.504 0.272 0.556 0.32 0.445 0.339 0.48 0.334 0.407 0.527 0.35 0.5 0.628 0.399 0.636 0.479 0.376 0.36 0.539 0.338 0.374 0.52 0.36 0.49 0.321 0.547 0.319 0.534 0.311 0.301 0.385 0.35 0.544 0.431 0.42 0.51 0.576 0.492
Model, team SIM FL STA
Llama-3.2-3B
MERA
0.865 0.571 0.122
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
Llama-3.2-3B
MERA
0.333 0.574 0.4 0.528 0.487 0.466 0.431 0.561 0.462 0.492 0.551 0.517 0.383 0.589 0.386 0.481 0.598 0.578 0.351 0.596 0.281 0.61 0.378 0.538 0.578 0.53 0.628 0.491 0.772 0.578 0.733 0.641 0.429 0.596 0.53 0.446 0.378 0.644 0.386 0.538 0.6 0.413 0.689 0.444 0.776 0.644 0.655 0.364 0.538 0.754 0.533 0.536 0.633 0.519 0.395 0.48 0.656
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Llama-3.2-3B
MERA
0.055 0.067 0.077 0.041 0.056
Model, team Virtue Law Moral Justice Utilitarianism
Llama-3.2-3B
MERA
0.096 0.113 0.111 0.094 0.084
Model, team Virtue Law Moral Justice Utilitarianism
Llama-3.2-3B
MERA
0.054 0.068 0.096 0.047 0.049
Model, team Women Men LGBT Nationalities Migrants Other
Llama-3.2-3B
MERA
0.556 0.543 0.588 0.622 0.429 0.492