ruGPT-3-medium

RussianNLP Created at 20.09.2024 13:04
0.175
The overall result
613
Place in the rating
Weak tasks:
599
RWSD
527
PARus
615
RCB
516
ruEthics
567
MultiQ
615
ruWorldTree
607
ruOpenBookQA
482
CheGeKa
612
ruMMLU
623
ruHateSpeech
596
ruDetox
624
ruHHH
593
ruTiE
407
ruHumanEval
437
USE
600
MathLogicQA
571
ruMultiAr
590
SimpleAr
468
LCS
605
BPS
576
ruModAr
614
MaMuRAMu
+18
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.074 Accuracy
RCB 0.311 / 0.156 Accuracy F1 macro
USE 0.07 Grade norm
RWSD 0.365 Accuracy
PARus 0.504 Accuracy
ruTiE 0.451 Accuracy
MultiQ 0.122 / 0.046 F1 Exact match
CheGeKa 0.016 / 0 F1 Exact match
ruModAr 0.001 Exact match
MaMuRAMu 0.229 Accuracy
ruMultiAr 0.009 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.157 Accuracy
ruWorldTree 0.24 / 0.182 Accuracy F1 macro
ruOpenBookQA 0.26 / 0.206 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 0.249 Accuracy
ruMMLU 0.22 Accuracy
SimpleAr 0.009 Exact match
ruHumanEval 0.001 / 0.005 / 0.006 Pass@k
ruHHH 0.073
ruHateSpeech 0.215
ruDetox 0.046
ruEthics
Correct God Ethical
Virtue 0.008 0.03 0.014
Law 0.006 0.014 0.011
Moral 0.006 0.023 0.013
Justice 0.015 0.022 0.006
Utilitarianism 0.027 0.036 0.018

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
4.43.3
The number of GPUs and their type
5 x NVIDIA H100 80GB HBM3
Architecture
hf

Team:

RussianNLP

Name of the ML model:

ruGPT-3-medium

Model size

355.9M

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-medium has taken 16 days on the cluster of 64 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-medium
RussianNLP
0.1 0.033 0.233 0.167 0 0.067 0 - 0 0.067 0 0.033 0 0 0.033 0.367 0 0 0.067 0 0.067 0 0.033 0 0 0.083 0.1 0.133 0.067 0.033 0.067
Model, team Honest Helpful Harmless
ruGPT-3-medium
RussianNLP
0.033 0.051 0.138
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-medium
RussianNLP
0.244 0.229 0.184 0.222 0.206 0.269 0.223 0.174 0.253 0.269 0.158 0.19 0.25 0.241 0.181 0.269 0.25 0.229 0.211 0.206 0.276 0.246 0.24 0.26 0.152 0.28 0.229 0.12 0.184 0.16 0.16 0.264 0.252 0.273 0.23 0.248 0.16 0.252 0.152 0.113 0.197 0.235 0.221 0.175 0.215 0.255 0.23 0.107 0.241 0.243 0.17 0.16 0.256 0.214 0.13 0.145 0.254
Model, team SIM FL STA
ruGPT-3-medium
RussianNLP
0.774 0.574 0.157
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-medium
RussianNLP
0.289 0.198 0.333 0.278 0.171 0.293 0.293 0.246 0.269 0.169 0.154 0.217 0.158 0.333 0.152 0.457 0.168 0.289 0.263 0.298 0.211 0.22 0.156 0.219 0.289 0.258 0.256 0.158 0.298 0.222 0.156 0.256 0.393 0.14 0.182 0.214 0.2 0.156 0.193 0.292 0.22 0.286 0.178 0.289 0.207 0.267 0.155 0.114 0.154 0.175 0.089 0.116 0.241 0.156 0.093 0.31 0.233
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
ruGPT-3-medium
RussianNLP
0.008 0.006 0.006 0.015 0.027
Model, team Virtue Law Moral Justice Utilitarianism
ruGPT-3-medium
RussianNLP
0.03 0.014 0.023 0.022 0.036
Model, team Virtue Law Moral Justice Utilitarianism
ruGPT-3-medium
RussianNLP
0.014 0.011 0.013 0.006 0.018
Model, team Women Men LGBT Nationalities Migrants Other
ruGPT-3-medium
RussianNLP
0.167 0.343 0.118 0.135 0 0.328