sergeyzh/rubert-tiny-turbo

Быстрая модель BERT для расчетов эмбеддингов предложений на русском языке. Модель основана на cointegrated/rubert-tiny2 - имеет аналогичные размеры контекста (2048), ембеддинга (312) и быстродействие.

Использование

from sentence_transformers import SentenceTransformer, util

model = SentenceTransformer('sergeyzh/rubert-tiny-turbo')

sentences = ["привет мир", "hello world", "здравствуй вселенная"]
embeddings = model.encode(sentences)
print(util.dot_score(embeddings, embeddings))

Метрики

Оценки модели на бенчмарке encodechka:

model	CPU	GPU	size	Mean S	Mean S+W	dim
sergeyzh/LaBSE-ru-turbo	120.40	8.05	490	0.789	0.702	768
BAAI/bge-m3	523.40	22.50	2166	0.787	0.696	1024
intfloat/multilingual-e5-large	506.80	30.80	2136	0.780	0.686	1024
intfloat/multilingual-e5-base	130.61	14.39	1061	0.761	0.669	768
sergeyzh/rubert-tiny-turbo	5.51	3.25	111	0.749	0.667	312
intfloat/multilingual-e5-small	40.86	12.09	449	0.742	0.645	384
cointegrated/rubert-tiny2	5.51	3.25	111	0.704	0.638	312

model	STS	PI	NLI	SA	TI	IA	IC	ICX	NE1	NE2
sergeyzh/LaBSE-ru-turbo	0.864	0.748	0.490	0.814	0.974	0.806	0.815	0.801	0.305	0.404
BAAI/bge-m3	0.864	0.749	0.510	0.819	0.973	0.792	0.809	0.783	0.240	0.422
intfloat/multilingual-e5-large	0.862	0.727	0.473	0.810	0.979	0.798	0.819	0.773	0.224	0.374
intfloat/multilingual-e5-base	0.835	0.704	0.459	0.796	0.964	0.783	0.802	0.738	0.235	0.376
sergeyzh/rubert-tiny-turbo	0.828	0.722	0.476	0.787	0.955	0.757	0.780	0.685	0.305	0.373
intfloat/multilingual-e5-small	0.822	0.714	0.457	0.758	0.957	0.761	0.779	0.691	0.234	0.275
cointegrated/rubert-tiny2	0.750	0.651	0.417	0.737	0.937	0.746	0.757	0.638	0.360	0.386

Оценки модели на бенчмарке ruMTEB:

Model Name	Metric	sbert_large_ mt_nlu_ru	sbert_large_ nlu_ru	rubert-tiny2	rubert-tiny-turbo	multilingual-e5-small	multilingual-e5-base	multilingual-e5-large
CEDRClassification	Accuracy	0.368	0.358	0.369	0.390	0.401	0.423	0.448
GeoreviewClassification	Accuracy	0.397	0.400	0.396	0.414	0.447	0.461	0.497
GeoreviewClusteringP2P	V-measure	0.584	0.590	0.442	0.597	0.586	0.545	0.605
HeadlineClassification	Accuracy	0.772	0.793	0.742	0.686	0.732	0.757	0.758
InappropriatenessClassification	Accuracy	0.646	0.625	0.586	0.591	0.592	0.588	0.616
KinopoiskClassification	Accuracy	0.503	0.495	0.491	0.505	0.500	0.509	0.566
RiaNewsRetrieval	NDCG@10	0.214	0.111	0.140	0.513	0.700	0.702	0.807
RuBQReranking	MAP@10	0.561	0.468	0.461	0.622	0.715	0.720	0.756
RuBQRetrieval	NDCG@10	0.298	0.124	0.109	0.517	0.685	0.696	0.741
RuReviewsClassification	Accuracy	0.589	0.583	0.570	0.607	0.612	0.630	0.653
RuSTSBenchmarkSTS	Pearson correlation	0.712	0.588	0.694	0.787	0.781	0.796	0.831
RuSciBenchGRNTIClassification	Accuracy	0.542	0.539	0.456	0.529	0.550	0.563	0.582
RuSciBenchGRNTIClusteringP2P	V-measure	0.522	0.504	0.414	0.481	0.511	0.516	0.520
RuSciBenchOECDClassification	Accuracy	0.438	0.430	0.355	0.415	0.427	0.423	0.445
RuSciBenchOECDClusteringP2P	V-measure	0.473	0.464	0.381	0.411	0.443	0.448	0.450
SensitiveTopicsClassification	Accuracy	0.285	0.280	0.220	0.244	0.228	0.234	0.257
TERRaClassification	Average Precision	0.520	0.502	0.519	0.563	0.551	0.550	0.584

Model Name	Metric	sbert_large_ mt_nlu_ru	sbert_large_ nlu_ru	rubert-tiny2	rubert-tiny-turbo	multilingual-e5-small	multilingual-e5-base	multilingual-e5-large
Classification	Accuracy	0.554	0.552	0.514	0.535	0.551	0.561	0.588
Clustering	V-measure	0.526	0.519	0.412	0.496	0.513	0.503	0.525
MultiLabelClassification	Accuracy	0.326	0.319	0.294	0.317	0.314	0.329	0.353
PairClassification	Average Precision	0.520	0.502	0.519	0.563	0.551	0.550	0.584
Reranking	MAP@10	0.561	0.468	0.461	0.622	0.715	0.720	0.756
Retrieval	NDCG@10	0.256	0.118	0.124	0.515	0.697	0.699	0.774
STS	Pearson correlation	0.712	0.588	0.694	0.787	0.781	0.796	0.831
Average	Average	0.494	0.438	0.431	0.548	0.588	0.594	0.630