PEG Open-source Text Retrieval Model - Progressive Learning, Precise Retrieval with Weights Adjusted by Negative Sample Difficulty

PEG

Developed by TownsWu

PEG is a model that achieves robust text retrieval through progressive learning, adjusting loss weights based on the difficulty levels of negative samples.

Text Embedding

Transformers

Chinese#Progressive Learning #Text Retrieval #Multi-domain Adaptation

Downloads 36

Release Time : 4/25/2025

Model Overview

The PEG model (Progressive Embedding for Text) gradually adjusts the weights of samples contributing to the loss in large batches based on the difficulty levels of negative samples. It is suitable for text retrieval tasks across a wide range of domains.

Model Features

Progressive Learning

Gradually adjusts loss weights based on the difficulty levels of negative samples to enhance model performance.

Large-scale Training Data

Trained on a dataset of over 110 million entries covering a wide range of domains.

Robust Text Retrieval

Performs excellently in various domains (e.g., general knowledge, finance, tourism, medicine, etc.).

Model Capabilities

Text embedding generation

Sentence similarity calculation

Text retrieval

Question-answering system support

Use Cases

Medical Field

Medical QA Retrieval

Used for retrieving related questions in medical question-answering systems.

Performs excellently on the CMedQAv1 and CMedQAv2 datasets.

General Retrieval

General Text Retrieval

Suitable for text retrieval tasks across various domains.

Performs well in multiple retrieval tasks.

🚀 PEG

PEG is a model designed for sentence similarity tasks, leveraging technologies such as feature extraction and transformers to achieve high performance in various retrieval and reranking tasks.

📚 Documentation

Model Information

Property	Details
Pipeline Tag	Sentence Similarity
Tags	PEG, Feature Extraction, Sentence Similarity, Transformers, MTEB
Model Name	PEG

Performance Metrics

Reranking Tasks

Task	Dataset	MAP	MRR
Reranking	MTEB CMedQAv1	84.09137463267582	86.6288888888889
Reranking	MTEB CMedQAv2	86.55765031914974	89.4325396825397
Reranking	MTEB MMarcoReranking	33.548800108363665	32.529761904761905
Reranking	MTEB T2Reranking	69.43381583724414	80.47879657392181

Retrieval Tasks

Task	Dataset	MAP@1	MAP@10	MAP@100	MAP@1000	MAP@3	MAP@5	MRR@1	MRR@10	MRR@100	MRR@1000	MRR@3	MRR@5	NDCG@1	NDCG@10	NDCG@100	NDCG@1000	NDCG@3	NDCG@5	Precision@1	Precision@10	Precision@100	Precision@1000	Precision@3	Precision@5	Recall@1	Recall@10	Recall@100	Recall@1000	Recall@3	Recall@5
Retrieval	MTEB CmedqaRetrieval	26.101000000000003	38.239000000000004	40.083	40.205	34.386	36.425999999999995	39.434999999999995	46.967999999999996	47.946	47.997	44.803	45.911	39.434999999999995	44.416	51.773	53.888000000000005	39.816	41.467999999999996	39.434999999999995	9.786999999999999	1.5810000000000002	0.184	22.414	15.943999999999999	26.101000000000003	53.82900000000001	84.63199999999999	98.782	39.585	45.141
Retrieval	MTEB CovidRetrieval	70.39	78.93599999999999	79.202	79.205	77.538	78.312	70.706	79.018	79.28399999999999	79.288	77.713	78.462	70.601	82.555	83.718	83.855	79.779	81.149	70.601	9.463000000000001	0.9979999999999999	0.101	28.871999999999996	18.019	70.39	93.572	98.736	99.895	86.091	89.384
Retrieval	MTEB DuRetrieval	26.147	80.205	82.96	82.999	55.16799999999999	69.798	89.8	93.16799999999999	93.22500000000001	93.228	92.85	93.067	89.8	87.668	90.16	90.505	85.842	85.101	89.8	42.225	4.8149999999999995	0.48900000000000005	76.967	65.32	26.147	89.30399999999999	97.609	99.409	57.56	74.78200000000001
Retrieval	MTEB EcomRetrieval	53.300000000000004	62.507000000000005	63.068000000000005	63.08200000000001	60.050000000000004	61.41	53.300000000000004	62.507000000000005	63.068000000000005	63.08200000000001	60.050000000000004	61.41	53.300000000000004	67.31700000000001	69.862	70.231	62.222	64.66300000000001	53.300000000000004	8.260000000000002	0.941	0.097	22.833000000000002	14.879999999999999	53.300000000000004	82.6	94.1	97.0	68.5	74.4
Retrieval	MTEB MMarcoRetrieval	70.68799999999999	79.28399999999999	79.537	79.545	77.643	78.694	73.05199999999999	79.794	80.024	80.03099999999999	78.441	79.29	73.05199999999999	82.627	83.737	83.946	79.585	81.306	73.05199999999999	9.835	1.038	0.106	29.756	18.788	70.68799999999999	92.38300000000001	97.347	98.992	84.37	88.434
Retrieval	MTEB MedicalRetrieval	53.1	58.36599999999999	58.939	58.99100000000001	57.15	57.794999999999995	53.2	58.416000000000004	58.989999999999995	59.041	57.199999999999996	57.845	53.1	60.989000000000004	63.967	65.436	58.425000000000004	59.583	53.1	6.93	0.8370000000000001	0.096	20.7	12.98	53.1	69.3	83.7	95.5	62.1	64.9
Retrieval	MTEB T2Retrieval	28.116000000000003	80.026	83.541	83.592	56.092	69.114	91.557	93.73700000000001	93.808	93.811	93.384	93.614	91.553	87.003	90.128	90.615	88.205	86.978	91.553	43.25	5.067	0.518	77.25	64.902	28.116000000000003	85.994	96.345	98.867	57.67099999999999	72.26
Retrieval	MTEB VideoRetrieval	64.9	73.763	74.116	74.12100000000001	72.15	73.25	64.9	...	...	...	...	...	64.9	...	...	...	...	...	64.9	...	...	...	...	...	64.9	...	...	...	...	...

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご