This repository contains a collection of domain-adapted embedding models trained for semantic search and information retrieval in a specialized scientific domain.
Each model was fine-tuned using a combination of domain-adaptive pretraining and contrastive sentence-level learning on a large corpus of scientific papers. The models are designed to produce high-quality text embeddings optimized for query–document matching tasks.
The collection includes both standalone fine-tuned models and models intended for use in ensemble configurations, where multiple encoders are combined to improve retrieval performance.
All models are compatible with standard Sentence Transformers / HuggingFace pipelines and can be used for embedding generation, semantic similarity computation, and downstream retrieval tasks.
