Qwen1.5-MoE-A2.7B-Chat

MERA Created at 22.09.2024 21:41

0.326

The overall result

391

Place in the rating

Weak tasks:

539

RWSD

404

PARus

401

RCB

401

MultiQ

339

ruWorldTree

304

ruOpenBookQA

338

CheGeKa

335

ruMMLU

385

ruHateSpeech

333

ruDetox

507

ruHHH

375

ruTiE

317

ruHumanEval

385

USE

359

MathLogicQA

293

ruMultiAr

253

SimpleAr

298

LCS

424

BPS

431

ruModAr

306

MaMuRAMu

+17

Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name	Result	Metric
LCS	0.094	Accuracy
RCB	0.429 / 0.401	Accuracy F1 macro
USE	0.069	Grade norm
RWSD	0.377	Accuracy
PARus	0.56	Accuracy
ruTiE	0.532	Accuracy
MultiQ	0.232 / 0.132	F1 Exact match
CheGeKa	0.038 / 0.019	F1 Exact match
ruModAr	0.136	Exact match
MaMuRAMu	0.547	Accuracy
ruMultiAr	0.22	Exact match
ruCodeEval	0 / 0 / 0	Pass@k
MathLogicQA	0.338	Accuracy
ruWorldTree	0.724 / 0.724	Accuracy F1 macro
ruOpenBookQA	0.665 / 0.665	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

0.596

Accuracy

ruMMLU

0.447

Accuracy

SimpleAr

0.938

Exact match

ruHumanEval

0.002 / 0.006 / 0.006

Pass@k

ruHHH

0.461

ruHateSpeech

0.551

ruDetox

0.125

ruEthics

	Correct	God	Ethical
Virtue	-0	0.013	-0.016
Law	-0.004	0.023	-0.036
Moral	0.009	0.007	-0.024
Justice	-0.017	-0.031	-0.03
Utilitarianism	-0.032	-0.009	0.01

Information about the submission:

Mera version

v.1.2.0

Torch Version

2.4.0

The version of the codebase

9b26db97

CUDA version

12.1

Precision of the model weights

bfloat16

Seed

1234

Batch

Transformers version

4.43.2

The number of GPUs and their type

2 x NVIDIA H100 80GB HBM3

Architecture

vllm

Team:

MERA

Name of the ML model:

Qwen1.5-MoE-A2.7B-Chat

Link to the ML model:

https://huggingface.co/Qwen/Qwen1.5-MoE-A2.7B-Chat

Model size

14.3B

Model type:

Opened

SFT

Additional links:

https://qwenlm.github.io/blog/qwen-moe/

Architecture description:

Qwen1.5-MoE is a transformer-based MoE decoder-only language model pretrained on a large amount of data.

Description of the training:

The model was pretrained with a large amount of data, and then it was post-trained with both supervised finetuning and direct preference optimization.

Pretrain data:

License:

tongyi-qianwen

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 - 32768 \nrutie - 2000 \nrutie - 5000

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.

Expand information

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
Qwen1.5-MoE-A2.7B-Chat MERA	0.3	0.1	0.233	0.167	0	0.2	0	-	0.033	0	0	0	0.033	0	0.067	0.317	0	0	0	0	0	0	0	0	0	0.125	0	0.033	0.033	0	0

Model, team	Honest	Helpful	Harmless
Qwen1.5-MoE-A2.7B-Chat MERA	0.557	0.424	0.397

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
Qwen1.5-MoE-A2.7B-Chat MERA	0.341	0.398	0.474	0.701	0.497	0.667	0.583	0.518	0.531	0.516	0.254	0.27	0.3	0.528	0.557	0.497	0.55	0.41	0.244	0.504	0.265	0.567	0.25	0.428	0.375	0.48	0.351	0.565	0.535	0.31	0.6	0.678	0.497	0.626	0.472	0.462	0.29	0.555	0.358	0.394	0.571	0.327	0.441	0.379	0.558	0.315	0.529	0.352	0.34	0.374	0.31	0.599	0.431	0.496	0.57	0.57	0.544

Model, team	SIM	FL	STA
Qwen1.5-MoE-A2.7B-Chat MERA	0.613	0.549	0.469

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
Qwen1.5-MoE-A2.7B-Chat MERA	0.378	0.554	0.383	0.583	0.553	0.621	0.414	0.561	0.635	0.492	0.628	0.492	0.425	0.581	0.48	0.407	0.579	0.511	0.316	0.579	0.263	0.678	0.444	0.503	0.578	0.576	0.654	0.474	0.842	0.644	0.733	0.564	0.446	0.684	0.439	0.571	0.267	0.778	0.404	0.477	0.584	0.556	0.644	0.4	0.793	0.778	0.707	0.568	0.646	0.737	0.756	0.609	0.595	0.519	0.349	0.444	0.689

Coorect

Good

Ethical

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
Qwen1.5-MoE-A2.7B-Chat MERA	-0	-0.004	0.009	-0.017	-0.032

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
Qwen1.5-MoE-A2.7B-Chat MERA	0.013	0.023	0.007	-0.031	-0.009

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
Qwen1.5-MoE-A2.7B-Chat MERA	-0.016	-0.036	-0.024	-0.03	0.01

Model, team	Women	Men	LGBT	Nationalities	Migrants	Other
Qwen1.5-MoE-A2.7B-Chat MERA	0.611	0.486	0.471	0.649	0.143	0.492

Qwen1.5-MoE-A2.7B-Chat

Ratings for leaderboard tasks

Evaluation on open tasks:

Information about the submission:

Team:

Name of the ML model:

Link to the ML model:

Model size

Model type:

Additional links:

Architecture description:

Description of the training:

Pretrain data:

License:

Inference parameters

Ratings by subcategory

Confirm the deletion of the sub