GigaChat-20B-A3B

GIGACHAT Created at 13.12.2024 07:11

0.513

The overall result

Place in the rating

In the top by tasks:

ruModAr

The task is one of the main ones

Weak tasks:

300

RWSD

178

PARus

230

RCB

179

ruEthics

182

MultiQ

187

ruWorldTree

155

ruOpenBookQA

CheGeKa

181

ruMMLU

123

ruHateSpeech

181

ruDetox

151

ruHHH

106

ruTiE

181

ruHumanEval

USE

146

MathLogicQA

184

ruMultiAr

277

SimpleAr

412

LCS

234

BPS

112

MaMuRAMu

153

ruCodeEval

+18

Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name	Result	Metric
LCS	0.07	Accuracy
RCB	0.518 / 0.441	Accuracy F1 macro
USE	0.334	Grade norm
RWSD	0.512	Accuracy
PARus	0.842	Accuracy
ruTiE	0.758	Accuracy
MultiQ	0.393 / 0.187	F1 Exact match
CheGeKa	0.318 / 0.252	F1 Exact match
ruModAr	0.87	Exact match
MaMuRAMu	0.741	Accuracy
ruMultiAr	0.272	Exact match
ruCodeEval	0.041 / 0.054 / 0.061	Pass@k
MathLogicQA	0.455	Accuracy
ruWorldTree	0.901 / 0.901	Accuracy F1 macro
ruOpenBookQA	0.833 / 0.833	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

0.921

Accuracy

ruMMLU

0.587

Accuracy

SimpleAr

0.923

Exact match

ruHumanEval

0.037 / 0.04 / 0.043

Pass@k

ruHHH

0.73

ruHateSpeech

0.777

ruDetox

0.191

ruEthics

	Correct	God	Ethical
Virtue	0.299	0.314	0.359
Law	0.337	0.329	0.37
Moral	0.351	0.334	0.403
Justice	0.251	0.266	0.327
Utilitarianism	0.249	0.295	0.329

Information about the submission:

Mera version

v.1.2.0

Torch Version

2.4.0

The version of the codebase

db539c9

CUDA version

12.1

Precision of the model weights

Seed

1234

Butch

Transformers version

4.46.0.dev0

The number of GPUs and their type

5 x NVIDIA H100 80GB HBM3

Architecture

gigachat_llms

Team:

GIGACHAT

Name of the ML model:

GigaChat-20B-A3B

Link to the ML model:

https://huggingface.co/ai-sage/GigaChat-20B-A3B-instruct

Model size

20.0B

Model type:

Opened

SFT

MoE

Architecture description:

GigaChat-20B-A3B is a Large Language Model (LLM) that was fine-tuned on instruction corpus and has context length of 32k tokens. GigaChat-20B-A3B is Mixture of Experts model and has 3.3B active parameters. The model is available at https://huggingface.co/ai-sage/GigaChat-20B-A3B-instruct as instruct version and https://huggingface.co/ai-sage/GigaChat-20B-A3B-base as base version

Description of the training:

Pretrain data:

License:

Open-source model by Sber

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;

Description of the template:
{% if messages[0]['role'] == 'system' -%}\n {%- set loop_messages = messages[1:] -%}\n {%- set system_message = bos_token + messages[0]['content'] + additional_special_tokens[1] -%}\n{%- else -%}\n {%- set loop_messages = messages -%}\n {%- set system_message = bos_token + '' -%}\n{%- endif -%}\n{%- for message in loop_messages %}\n {% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}\n {{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}\n {% endif %}\n \n {%- if loop.index0 == 0 -%}\n {{ system_message -}}\n {%- endif -%}\n {%- if message['role'] == 'user' -%}\n {{ message['role'] + additional_special_tokens[0] + message['content'] + additional_special_tokens[1] -}}\n {{ 'available functions' + additional_special_tokens[0] + additional_special_tokens[2] + additional_special_tokens[3] + additional_special_tokens[1] -}}\n {%- endif -%}\n {%- if message['role'] == 'assistant' -%}\n {{ message['role'] + additional_special_tokens[0] + message['content'] + additional_special_tokens[1] -}}\n {%- endif -%}\n {%- if loop.last and add_generation_prompt -%}\n {{ 'assistant' + additional_special_tokens[0] -}}\n {%- endif -%}\n{%- endfor %}

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
GigaChat-20B-A3B GIGACHAT	0.9	0.233	0.767	0.2	0.133	0.433	0.167	-	0.133	0.167	0.167	0.133	0.533	0.133	0.067	0.567	0	0.067	0.033	0.033	0.067	0.567	0.433	0.033	0.033	0.7	0.333	0.433	0.433	0.3	0.5

Model, team	Honest	Helpful	Harmless
GigaChat-20B-A3B GIGACHAT	0.689	0.797	0.707

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
GigaChat-20B-A3B GIGACHAT	0.533	0.476	0.697	0.825	0.706	0.811	0.728	0.659	0.664	0.61	0.377	0.476	0.36	0.713	0.764	0.618	0.63	0.667	0.389	0.702	0.316	0.772	0.37	0.584	0.384	0.65	0.403	0.62	0.714	0.41	0.67	0.769	0.663	0.717	0.675	0.513	0.39	0.752	0.371	0.488	0.793	0.621	0.586	0.454	0.792	0.481	0.779	0.367	0.404	0.582	0.47	0.768	0.659	0.664	0.69	0.739	0.798

Model, team	SIM	FL	STA
GigaChat-20B-A3B GIGACHAT	0.35	0.783	0.748

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
GigaChat-20B-A3B GIGACHAT	0.511	0.822	0.617	0.62	0.816	0.845	0.638	0.719	0.788	0.646	0.756	0.692	0.475	0.822	0.731	0.765	0.71	0.644	0.667	0.737	0.246	0.78	0.689	0.787	0.711	0.803	0.821	0.684	0.825	0.733	0.844	0.833	0.723	0.912	0.667	0.732	0.667	0.822	0.614	0.692	0.857	0.841	0.778	0.689	0.879	0.867	0.914	0.727	0.831	0.877	0.756	0.754	0.81	0.662	0.465	0.743	0.789

Coorect

Good

Ethical

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
GigaChat-20B-A3B GIGACHAT	0.299	0.337	0.351	0.251	0.249

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
GigaChat-20B-A3B GIGACHAT	0.314	0.329	0.334	0.266	0.295

Model, team	Virtue	Law	Moral	Justice	Utilitarianism
GigaChat-20B-A3B GIGACHAT	0.359	0.37	0.403	0.327	0.329

Model, team	Women	Men	LGBT	Nationalities	Migrants	Other
GigaChat-20B-A3B GIGACHAT	0.815	0.629	0.647	0.757	0.714	0.852

GigaChat-20B-A3B

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:

Architecture description:

Description of the training:

Pretrain data:

License:

Inference parameters

Ratings by subcategory

Confirm the deletion of the sub