starred/minne
1
0
Fork 0
mirror of https://github.com/perstarkse/minne.git synced 2026-04-23 17:28:34 +02:00
Code Issues Packages Projects Releases 10 Wiki Activity

release: 1.0.0

Browse Source
This commit is contained in:
Per Stark
2026-01-11 18:37:07 +01:00
parent db43be1606
commit 8fe4ac9fec
53 changed files with 757 additions and 630 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
CHANGELOG.md
View File

@@ -1,8 +1,9 @@
# Changelog
## Unreleased
## 1.0.0 (2026-01-02)
- **Locally generated embeddings are now default**. If you want to continue using API embeddings, set EMBEDDING_BACKEND to openai. This will download a ONNX model and recreate all embeddings. But in most instances it's very worth it. Removing the network bound call to create embeddings. Creating embeddings on my N100 device is extremely fast. Typically a search response is provided in less than 50ms.
- Added a benchmarks create for evaluating the retrieval process
- Added fastembed embedding support, enables the use of local CPU generated embeddings, greatly improved latency if machine can handle it. Quick search has vastly better accuracy and is much faster, 50ms latency when testing compared to minimum 300ms.
- Embeddings stored on own table
- Embeddings stored on own table.
- Refactored retrieval pipeline to use the new, faster and more accurate strategy. Read [blog post](https://blog.stark.pub/posts/eval-retrieval-refactor/) for more details.
## Version 0.2.7 (2025-12-04)