MedCPT-Query-Encoder is a biomedical text embedding model hosted on Hugging Face. It belongs to the MedCPT suite released by the NLM/DIR BioNLP Group at NCBI and converts queries into vector representations for semantic retrieval applications.
The model is provided as a BERT-based transformer for feature extraction. It is distributed in PyTorch and Safetensors formats and supports text-embeddings-inference. Users load it through the Transformers library with either a high-level pipeline for feature extraction or direct instantiation of the AutoTokenizer and AutoModel classes. The repository includes example code for both approaches and references a Google Colab notebook for demonstration.
It carries a public-domain license. The underlying research appears in arXiv preprint 2307.00589. The page lists an introduction to MedCPT along with usage cases for the query encoder and semantic search over PubMed, though implementation details for those cases reside in linked notebooks rather than the model card itself.
The model targets developers and researchers integrating embedding-based search into biomedical systems. It is delivered exclusively as a downloadable model artifact on the Hugging Face platform.
MedCPT Query Encoder sits in PulseGate's Foundation models & chat category. It focuses on finding relevant biomedical literature by encoding queries into dense vector representations. MedCPT Query Encoder is an open-source project aimed at researchers. The project is open source (Open Source). MedCPT Query Encoder is available on the web and API.
Behind MedCPT Query Encoder is NCBI, based in the United States, and the product first shipped in 2023. Among its 3 catalogued features are Text Embeddings, Semantic Search, and Biomedical Domain. It exposes integrations via a public API.
Latest indexed changes and source events
ncbi/MedCPT-Query-Encoder verified by the PulseGate indexer
Other apps tracked under the same category.
