fix-normalization-after-truncate
#53
by
jupyterjazz
- opened
We had a bug in the encode function, doing Matryoshka embedding truncation after normalization, resulting in non-normalized truncated embeddings:
https://huggingface.co/jinaai/jina-embeddings-v3/discussions/60
This PR fixes the issue by moving normalization after truncation
jupyterjazz
changed pull request status to
open
The changes look good 🚀 thanks for making the fix!
jupyterjazz
changed pull request status to
merged