ruadapt llama3-8B-instruct lep ft

RCC MSU Created at 25.09.2024 18:20
0.451
The overall result
237
Place in the rating
Weak tasks:
77
RWSD
212
PARus
147
RCB
138
ruEthics
148
MultiQ
285
ruWorldTree
265
ruOpenBookQA
206
CheGeKa
299
ruMMLU
393
ruHateSpeech
128
ruDetox
225
ruHHH
193
ruTiE
274
ruHumanEval
316
USE
355
MathLogicQA
302
ruMultiAr
275
SimpleAr
109
LCS
240
BPS
308
ruModAr
275
MaMuRAMu
+18
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.152 Accuracy
RCB 0.555 / 0.472 Accuracy F1 macro
USE 0.121 Grade norm
RWSD 0.619 Accuracy
PARus 0.842 Accuracy
ruTiE 0.72 Accuracy
MultiQ 0.459 / 0.309 F1 Exact match
CheGeKa 0.149 / 0.103 F1 Exact match
ruModAr 0.454 Exact match
MaMuRAMu 0.633 Accuracy
ruMultiAr 0.242 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.362 Accuracy
ruWorldTree 0.838 / 0.837 Accuracy F1 macro
ruOpenBookQA 0.765 / 0.764 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 0.943 Accuracy
ruMMLU 0.518 Accuracy
SimpleAr 0.943 Exact match
ruHumanEval 0.009 / 0.012 / 0.012 Pass@k
ruHHH 0.708
ruHateSpeech 0.589
ruDetox 0.262
ruEthics
Correct God Ethical
Virtue 0.339 0.381 0.361
Law 0.349 0.363 0.333
Moral 0.363 0.382 0.382
Justice 0.305 0.321 0.315
Utilitarianism 0.32 0.335 0.322

Information about the submission

Mera version
v.1.2.0
Torch Version
2.3.0
The version of the codebase
430295f
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
NVIDIA A100
Architecture
hf

Team:

RCC MSU

Name of the ML model:

ruadapt llama3-8B-instruct lep ft

Model size

8.4B

Model type:

Opened

SFT

Architecture description:

LoRa tuned version of ruadapt llama 3 8B with extended tokenizer after LEP (Learned Embedding Propagation, paper will be soon) procedure on saiga_scored d7 dataset. Thanks to the extended tokenizer, the model works more efficiently with the Russian language. How to cite: Tikhomirov M., Chernyshev D. Facilitating large language model Russian adaptation with Learned Embedding Propagation // 2024 (will be soon) Tikhomirov M., Chernyshev D. Impact of Tokenization on LLaMa Russian Adaptation //2023 Ivannikov Ispras Open Conference (ISPRAS). – IEEE, 2023. – С. 163-168.

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 - 8192

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.

Description of the template:
{% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|> \n \n'+ message['content'] | trim + '<|eot_id|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{% if add_generation_prompt %}{{ '<|start_header_id|>assistant<|end_header_id|> \n \n' }}{% endif %}

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
ruadapt llama3-8B-instruct lep ft
RCC MSU
0.167 0.133 0.733 0.267 0 0.1 0 - 0 0 0.067 0.067 0.1 0 0.133 0.267 0 0.033 0 0.033 0 0.2 0.067 0 0.067 0.1 0.033 0.067 0.467 0.167 0.267
Model, team Honest Helpful Harmless
ruadapt llama3-8B-instruct lep ft
RCC MSU
0.639 0.695 0.793
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
ruadapt llama3-8B-instruct lep ft
RCC MSU
0.496 0.416 0.651 0.718 0.598 0.721 0.641 0.608 0.552 0.583 0.377 0.357 0.34 0.63 0.635 0.59 0.63 0.521 0.333 0.634 0.26 0.713 0.31 0.439 0.402 0.55 0.379 0.583 0.694 0.37 0.62 0.752 0.479 0.707 0.574 0.517 0.33 0.677 0.404 0.458 0.672 0.441 0.628 0.406 0.654 0.384 0.652 0.352 0.365 0.5 0.47 0.743 0.515 0.55 0.69 0.703 0.674
Model, team SIM FL STA
ruadapt llama3-8B-instruct lep ft
RCC MSU
0.647 0.659 0.686
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
ruadapt llama3-8B-instruct lep ft
RCC MSU
0.489 0.634 0.6 0.565 0.605 0.69 0.483 0.596 0.692 0.6 0.654 0.65 0.483 0.659 0.608 0.667 0.682 0.644 0.491 0.702 0.298 0.712 0.511 0.657 0.622 0.667 0.615 0.579 0.825 0.689 0.778 0.692 0.589 0.789 0.47 0.536 0.511 0.733 0.456 0.677 0.682 0.587 0.822 0.622 0.879 0.778 0.81 0.591 0.615 0.807 0.8 0.667 0.684 0.494 0.488 0.573 0.744
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
ruadapt llama3-8B-instruct lep ft
RCC MSU
0.339 0.349 0.363 0.305 0.32
Model, team Virtue Law Moral Justice Utilitarianism
ruadapt llama3-8B-instruct lep ft
RCC MSU
0.381 0.363 0.382 0.321 0.335
Model, team Virtue Law Moral Justice Utilitarianism
ruadapt llama3-8B-instruct lep ft
RCC MSU
0.361 0.333 0.382 0.315 0.322
Model, team Women Men LGBT Nationalities Migrants Other
ruadapt llama3-8B-instruct lep ft
RCC MSU
0.602 0.457 0.647 0.676 0.714 0.557