julep-ai/julep

Upgrade embedding model from bge-m3 to gte-Qwen2-1.5B-instruct

Opened this issue · 0 comments

creatorrr commented

See gte-Qwen2-1.5B-instruct -- it's both lighter-weight and more performant but key differences are:

  • embedding dimensions: 1536 (instead of 1024)
  • uses an instruction for the queries.