lightblue/suzume-llama-3-8B-multilingual

BODBE LLM Created at 08.05.2024 13:29

0.453

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	0.142	Accuracy
RCB	0.521 / 0.424	Accuracy F1 macro
USE	0.018	Grade norm
RWSD	0.569	Accuracy
PARus	0.744	Accuracy
ruTiE	0.614	Accuracy
MultiQ	0.261 / 0.161	F1 Exact match
CheGeKa	0.035 / 0	F1 Exact match
ruModAr	0.59	Exact match
ruMultiAr	0.254	Exact match
MathLogicQA	0.373	Accuracy
ruWorldTree	0.844 / 0.844	Accuracy F1 macro
ruOpenBookQA	0.795 / 0.795	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

0.336

Accuracy

ruMMLU

0.712

Accuracy

SimpleAr

0.955

Exact match

ruHumanEval

0.01 / 0.052 / 0.104

Pass@k

ruHHH

0.663

ruHateSpeech

0.725

ruDetox

0.138

ruEthics

	Correct	God	Ethical
Virtue	-0.287	-0.327	-0.338
Law	-0.309	-0.314	-0.318
Moral	-0.283	-0.337	-0.356
Justice	-0.261	-0.273	-0.305
Utilitarianism	-0.227	-0.25	-0.274

Information about the submission:

Mera version

Torch Version

The version of the codebase

CUDA version

Precision of the model weights

Seed

Butch

Transformers version

The number of GPUs and their type

Architecture

Team:

BODBE LLM

Name of the ML model:

lightblue/suzume-llama-3-8B-multilingual

Link to the ML model:

https://huggingface.co/lightblue/suzume-llama-3-8B-multilingual

Architecture description:

Suzume 8B, многоязычная дообученная версия Llama 3 (meta-llama/Meta-Llama-3-8B-Instruct).

Description of the training:

Suzume 8B была дообучена Llama 3 на основе почти 90,000 многоязычных разговоров, что означает, что эта модель обладает интеллектом Llama 3, но дополнительно умеет общаться на большем количестве языков.

Pretrain data:

Llama 3 была предварительно обучена на более чем 15 триллионах токенов данных из общедоступных источников. Данные для дообучения включают общедоступные наборы инструкций, а также более 10 миллионов примеров с аннотациями от людей. Ни предварительные данные, ни данные для дообучения не включают данные пользователей Meta. Актуальность данных: марта 2023 г.

Training Details:

Эта модель была обучена с использованием 4 x A100 (80GB) в течение примерно 2.5 часов. Во время обучения использовались следующие гиперпараметры: learning_rate: 1e-05 train_batch_size: 2 eval_batch_size: 2 seed: 42 distributed_type: multi-GPU num_devices: 4 gradient_accumulation_steps: 2 total_train_batch_size: 16 total_eval_batch_size: 8 optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 lr_scheduler_type: cosine lr_scheduler_warmup_steps: 10 num_epochs: 1

License:

license: other license_name: llama-3 license_link: https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct/raw/main/LICENSE

Strategy, generation and parameters:

PyTorch version: 2.2.1+CUDA 12.1 Transformers: 4.40.1 lm-harness: v1.1.0 GPU: NVIDIA A100-SXM4-80GB

Expand information

Ratings by subcategory

Metric: Accuracy

Model, team	Honest	Helpful	Harmless
lightblue/suzume-llama-3-8B-multilingual BODBE LLM	0.59	0.593	0.81

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
lightblue/suzume-llama-3-8B-multilingual BODBE LLM	0.7	0.875	0.9	0.571	0.762	0.8	0.8	0.765	0.9	0.9	0.636	0.8	0.6	0.692	0.455	0.4	0.8	0.889	0.4	0.9	0.4	0.731	0.8	0.725	0.6	0.636	0.75	0.786	0.9	0.636	0.4	0.778	0.8	0.8	0.818	0.9	0.8	0.857	0.7	0.6	0.785	0.7	0.8	0.7	0.813	0.4	0.9	0.7	0.5	0.7	0.5	0.813	0.853	1	0.417	0.394	0.704

Model, team	SIM	FL	STA
lightblue/suzume-llama-3-8B-multilingual BODBE LLM	0.513	0.701	0.311

Coorect

Good

Ethical

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
lightblue/suzume-llama-3-8B-multilingual BODBE LLM	-0.287	-0.309	-0.283	-0.261	-0.227

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
lightblue/suzume-llama-3-8B-multilingual BODBE LLM	-0.327	-0.314	-0.337	-0.273	-0.25

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
lightblue/suzume-llama-3-8B-multilingual BODBE LLM	-0.338	-0.318	-0.356	-0.305	-0.274

Model, team	Women	Men	LGBT	Nationalities	Migrants	Other
lightblue/suzume-llama-3-8B-multilingual BODBE LLM	0.759	0.743	0.588	0.595	0.429	0.803

lightblue/suzume-llama-3-8B-multilingual

Ratings for leaderboard tasks

Evaluation on open tasks:

Information about the submission:

Team:

Name of the ML model:

Link to the ML model:

Architecture description:

Description of the training:

Pretrain data:

Training Details:

License:

Strategy, generation and parameters:

Ratings by subcategory

Confirm the deletion of the sub