abhinand/MedEmbed-small-v0.1 (ONNX, medical text embeddings)

#21
by krixi-dev - opened

Adds an ONNX version of abhinand/MedEmbed-small-v0.1, a BERT-based embedding model fine-tuned for medical and clinical text.

Why we want it
We run a healthcare social platform . Our posts are full of medical terms that general-purpose models handle unevenly. We tested this model with Typesense auto-embedding on 2,000 real surgical posts, and it returns relevant results for medical queries

Details

  • Source model: abhinand/MedEmbed-small-v0.1 (Apache 2.0)
  • Embedding size: 384
  • model_type: bert
  • Converted with: optimum-cli export onnx --model abhinand/MedEmbed-small-v0.1 --task feature-extraction
  • The export includes the sentence-embedding output (pooling inside the ONNX graph)
  • model_md5 and vocab_md5 in config.json are computed from the files in this PR
  • No indexing_prefix or query_prefix is needed
krixi-dev changed pull request title from Add abhinand/MedEmbed-small-v0.1 (ONNX, medical text embeddings) to abhinand/MedEmbed-small-v0.1 (ONNX, medical text embeddings)
krixi-dev changed pull request status to closed
krixi-dev changed pull request status to open

Hi. @Nithish2410 , @harisarang , @jasonbosco , @kishorenc : Could you kindly help review this merge request and approve it. Appreciate your help.

Ready to merge
This branch is ready to get merged automatically.

Sign up or log in to comment