[EXPERIMENTAL]
The objective of this repo is to build a blazing fast semantic Neural search using a multilingual LLM and Qdrant (Quadrant) vector db. The following techs are used
- Qdrant
- FastEmbed
- Sentence Transformer
- distiluse-base-multilingual-cased-v1 (model)
- generates aligned vector spaces, i.e., similar inputs in different languages are mapped close in vector space
- 14 languages (incl. NL)
- Scalar Quantization (for faster inference time and memory efficiency)
- 8-bit
The data was scraped by utilizing the NRC scraper api given a set of categories. The result is a data set of NRC article items > 6500 and chunks from full articles > 45000.
pip install -r requirements.txt
qdrant
http://localhost:6333
api
http://localhost:8000
web ui
index.html
swagger
http://localhost:8000/docs