mGPT 1.3B

MERA Created at 12.01.2024 11:17

0.198

The overall result

The submission does not contain all the required tasks

Ratings for leaderboard tasks

The table will scroll to the left

Task name	Result	Metric
LCS	0.136	Accuracy
RCB	0.333 / 0.167	Accuracy F1 macro
USE	0	Grade norm
RWSD	0.519	Accuracy
PARus	0.498	Accuracy
ruTiE	0.5	Accuracy
MultiQ	0.055 / 0.014	F1 Exact match
CheGeKa	0.004 / 0	F1 Exact match
ruModAr	0.001	Exact match
ruMultiAr	0.012	Exact match
MathLogicQA	0.258	Accuracy
ruWorldTree	0.251 / 0.225	Accuracy F1 macro
ruOpenBookQA	0.245 / 0.193	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

0.449

Accuracy

ruMMLU

0.241

Accuracy

SimpleAr

0.007

Exact match

ruHumanEval

0 / 0 / 0

Pass@k

ruHHH

0.478

ruHateSpeech

0.543

ruDetox

0.35

ruEthics

	Correct	God	Ethical
Virtue	0.071	0.055	0.03
Law	0.083	0.074	0.051
Moral	0.092	0.079	0.053
Justice	0.075	0.075	0.046
Utilitarianism	0.12	0.085	0.075

Information about the submission

Mera version

Torch Version

The version of the codebase

CUDA version

Precision of the model weights

Seed

Batch

Transformers version

The number of GPUs and their type

Architecture

Team:

MERA

Name of the ML model:

mGPT 1.3B

Link to the ML model:

https://huggingface.co/ai-forever/mGPT

Additional links:

https://arxiv.org/pdf/2204.07580.pdf

Architecture description:

The mGPT architecture is based on GPT-3. We use the architecture description by Brown et al., the GPT-2 code base (Radford et al., 2019) from HuggingFace (Wolf et al., 2020) and MegatronLM (Shoeybi et al., 2020). With all the other hyperparameters equal, GPT-3 has fewer layers (Layers: 48 vs. 24) but a larger hidden size (dmodel: 1600 vs. 2048) as opposed to GPT-2. GPT-3 also alternates the classic dense and sparse attention layers (Child et al., 2019).

Description of the training:

LM was pretrained with a total batch size of 2048 and a context window of 512 tokens. The total number of the training steps is 600k, and the models have seen 400B tokens during pretraining. The pretraining took 22 days on a cluster of 512 V100 GPUs for mGPT13B.

Pretrain data:

The pretraining corpus represents a collection of documents from Wikipedia and C4. The Wikipedia texts are extracted from the dumps (v. 20201101) with WikiExtractor (Attardi, 2015). The C4 data is downloaded using the Tensorflow datasets(Paper, 2021).

Training Details:

Fixed hyperparameters: vocabulary size of 100k, learning rate of 2e−4, and batch size of 4.

License:

MIT

Strategy, generation and parameters:

Code version v.1.1.0 All the parameters were not changed and are used as prepared by the organizers. Details: - 1 x NVIDIA A100 - dtype auto - Pytorch 2.1.2 + CUDA 12.1 - Transformers 4.36.2 - Context length 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
mGPT 1.3B MERA	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-

Model, team	Honest	Helpful	Harmless
mGPT 1.3B MERA	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
mGPT 1.3B MERA	0.4	0.125	0.3	0.257	0.19	0.3	0.2	0.118	0.2	0.2	0.273	0.7	0.4	0.346	0.273	0.3	0.2	0.222	0.4	0.4	0.3	0.192	0.2	0.216	0.2	0.273	0.313	0.143	0.4	0.273	0.2	0.389	0.2	0.3	0.455	0.2	0.1	0.143	0.3	0.2	0.215	0.2	0.3	0.2	0.313	0.2	0.1	0.2	0.3	0.2	0.182	0.5	0.324	0	0.167	0.182	0.111

Model, team	SIM	FL	STA
mGPT 1.3B MERA	0.721	0.601	0.734

Coorect

Good

Ethical

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
mGPT 1.3B MERA	0.071	0.083	0.092	0.075	0.12

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
mGPT 1.3B MERA	0.055	0.074	0.079	0.075	0.085

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
mGPT 1.3B MERA	0.03	0.051	0.053	0.046	0.075

Model, team	Women	Men	LGBT	Nationalities	Migrants	Other
mGPT 1.3B MERA	0.519	0.686	0.588	0.595	0.286	0.492

mGPT 1.3B

Ratings for leaderboard tasks

Evaluation on open tasks:

Information about the submission

Team:

Name of the ML model:

Link to the ML model:

Additional links:

Architecture description:

Description of the training:

Pretrain data:

Training Details:

License:

Strategy, generation and parameters:

Ratings by subcategory

Confirm the deletion of the sub