ruGPT-3-small

RussianNLP Created at 20.09.2024 12:09
0.203
The overall result
591
Place in the rating
Weak tasks:
525
RWSD
582
PARus
555
RCB
594
MultiQ
613
ruWorldTree
603
ruOpenBookQA
555
CheGeKa
596
ruMMLU
454
ruHateSpeech
584
ruDetox
535
ruHHH
529
ruTiE
391
USE
555
MathLogicQA
587
ruMultiAr
601
SimpleAr
304
LCS
517
BPS
571
ruModAr
604
MaMuRAMu
+16
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.104 Accuracy
RCB 0.347 / 0.345 Accuracy F1 macro
USE 0.081 Grade norm
RWSD 0.462 Accuracy
PARus 0.482 Accuracy
ruTiE 0.504 Accuracy
MultiQ 0.077 / 0.022 F1 Exact match
CheGeKa 0.008 / 0 F1 Exact match
ruModAr 0.001 Exact match
MaMuRAMu 0.259 Accuracy
ruMultiAr 0.003 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.263 Accuracy
ruWorldTree 0.242 / 0.226 Accuracy F1 macro
ruOpenBookQA 0.26 / 0.237 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 0.539 Accuracy
ruMMLU 0.255 Accuracy
SimpleAr 0.001 Exact match
ruHumanEval 0 / 0 / 0 Pass@k
ruHHH 0.478
ruHateSpeech 0.543
ruDetox 0.05
ruEthics
Correct God Ethical
Virtue -0.023 -0.019 -0.03
Law -0.022 -0.018 -0.028
Moral -0.024 -0.02 -0.032
Justice -0.021 -0.017 -0.027
Utilitarianism -0.028 -0.023 -0.036

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
float32
Seed
1234
Batch
1
Transformers version
4.42.3
The number of GPUs and their type
5 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

RussianNLP

Name of the ML model:

ruGPT-3-small

Model size

125.2M

Model type:

Opened

Pretrain

Architecture description:

ruGPT-3 is a Russian counterpart of GPT-3 (Brown et al., 2020). The model architecture is created using the description by Brown et al. and the GPT-2 code base (Radford et al., 2019) from the Transformers library. ruGPT-3 is pretrained on the language modeling objective. The BBPE tokenizer with the vocabulary size of 5 · 104 tokens was used.

Description of the training:

The model was pretrained with sequence length 1024 using transformers lib by the SberDevices team on 80B tokens for 3 epochs. After that, the model was finetuned 1 epoch with sequence length 2048. Total training time for ruGPT3-small has taken 7 days on the cluster of 32 V100-SXM3 GPUs. The initial learning rate of 1e−4 and the Adam optimizer with β1 = 0.9, β2 = 0.99, and ϵ = 1e−8 were used.

Pretrain data:

450GB of texts. The corpus includes texts from various publicly available resources, which represent diverse domains: Wikipedia, News, Books, Colossal Clean Crawled Corpus, OpenSubtitles.

License:

MIT

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 - 2048

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
ruGPT-3-small
RussianNLP
0.167 0 0.367 0 0 0 0 - 0.067 0.033 0 0.033 0.033 0 0.1 0.367 0 0 0.1 0 0.067 0 0.067 0 0 0.083 0.233 0.167 0.033 0.1 0.133
Model, team Honest Helpful Harmless
ruGPT-3-small
RussianNLP
0.492 0.475 0.466
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
ruGPT-3-small
RussianNLP
0.304 0.229 0.178 0.248 0.261 0.259 0.214 0.222 0.259 0.233 0.219 0.302 0.25 0.278 0.286 0.28 0.3 0.271 0.233 0.244 0.28 0.24 0.27 0.272 0.214 0.29 0.232 0.167 0.196 0.23 0.28 0.256 0.221 0.303 0.279 0.261 0.28 0.265 0.225 0.246 0.232 0.449 0.255 0.233 0.221 0.356 0.24 0.233 0.259 0.241 0.26 0.219 0.262 0.261 0.22 0.206 0.295
Model, team SIM FL STA
ruGPT-3-small
RussianNLP
0.715 0.584 0.202
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
ruGPT-3-small
RussianNLP
0.244 0.218 0.233 0.287 0.184 0.241 0.31 0.228 0.25 0.185 0.205 0.275 0.2 0.233 0.181 0.494 0.252 0.311 0.351 0.228 0.193 0.237 0.289 0.207 0.267 0.258 0.205 0.193 0.316 0.178 0.356 0.346 0.402 0.246 0.227 0.357 0.311 0.311 0.263 0.385 0.265 0.254 0.2 0.2 0.19 0.378 0.207 0.227 0.262 0.316 0.244 0.333 0.266 0.156 0.233 0.269 0.256
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
ruGPT-3-small
RussianNLP
-0.023 -0.022 -0.024 -0.021 -0.028
Model, team Virtue Law Moral Justice Utilitarianism
ruGPT-3-small
RussianNLP
-0.019 -0.018 -0.02 -0.017 -0.023
Model, team Virtue Law Moral Justice Utilitarianism
ruGPT-3-small
RussianNLP
-0.03 -0.028 -0.032 -0.027 -0.036
Model, team Women Men LGBT Nationalities Migrants Other
ruGPT-3-small
RussianNLP
0.509 0.686 0.588 0.595 0.286 0.508