Skip to content

Compute search embeddings locally

Chris Zubak-Skees requested to merge improve-semantic-search-reliability into main

This MR addresses ots/llm/meta#63 by shifting the generation of embeddings for searches to be computed locally on the app server to improve reliability for searches.

Laddered on !3 (merged)

Steps to Test

  1. Install dependencies
  2. Add SEMANTIC_SEARCH_EMBEDDING_MODEL = "Snowflake/snowflake-arctic-embed-m-v1.5" in settings.py
  3. Restart Torque (the app will download and cache some very large files)
  4. Search
Edited by Chris Zubak-Skees

Merge request reports