Skip to content

v1.9.2: Fix model alias resolution

Choose a tag to compare

View all tags
@knowusuboaky knowusuboaky released this 16 May 21:11
· 81 commits to main since this release
v1.9.2
55235f0

Bug Fixes

Fixed model alias resolution for higher-quality models (bge-base, bge-reranker, colbert-v2).

Problem

Models like bge-base, bge-reranker, and colbert-v2 weren't being recognized as embedded ONNX models, causing a fallback to sentence-transformers which isn't available on Databricks serverless.

Solution

  • Added new model folder names to _BUNDLED_MODEL_FOLDERS in easy.py
  • Added bge_base_en to BUNDLED_MODELS in DenseEmbedder for proper error handling
  • Added dimension lookup for 768-dim models (bge-base-en-v1.5)
  • Improved error messages for GitHub-hosted models with caching instructions

Verified

  • All 115 unit tests passing

Install

pip install vectrixdb==1.9.2

For Databricks

Cache models to a Unity Catalog Volume before running:

MODELS_VOLUME_PATH = '/Volumes/catalog/schema/vectrixdb_models/'
os.environ['VECTRIXDB_MODELS_DIR'] = MODELS_VOLUME_PATH
Assets 2
Loading