apunkt/sqlite-vec
1
0
Fork 0
mirror of https://github.com/asg017/sqlite-vec.git synced 2026-04-25 00:36:56 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

docs

Browse source
This commit is contained in:
Alex Garcia 2024-07-31 12:55:03 -07:00
parent 4febdff11a
commit 356f75cca7
17 changed files with 350 additions and 166 deletions
Download patch file Download diff file Expand all files Collapse all files

24
site/guides/matryoshka.md
View file

@ -2,17 +2,17 @@
Matryoshka embeddings are a new class of embedding models introduced in the
TODO-YYY paper [_TODO title_](https://arxiv.org/abs/2205.13147). They allow one
to truncate excess dimensions in large vector, without lossing much quality.
to truncate excess dimensions in large vector, without sacrificing much quality.
Let's say your embedding model generate 1024-dimensional vectors. If you have 1
million of these 1024-dimensional vectors, they would take up `4.096 GB` of
space! You're not able to reduce the dimensions without lossing a lot of
space! You're not able to reduce the dimensions without losing a lot of
quality - if you were to remove half of the dimensions 512-dimensional vectors,
you could expect to also lose 50% or more of the quality of results. There are
other dimensional-reduction techniques, like [PCA](#TODO), but this requires a
complicated and expensive training process.
other dimensional-reduction techniques, like [PCA](#TODO) or [Product Quantization](#TODO), but they typically require
complicated and expensive training processes.
Matryoshka embeddings, on the other hand, _can_ be truncated, without losing
Matryoshka embeddings, on the other hand, _can_ be truncated, without losing much
quality. Using [`mixedbread.ai`](#TODO) `mxbai-embed-large-v1` model, they claim
that
@ -20,16 +20,20 @@ They are called "Matryoshka" embeddings because ... TODO
## Matryoshka Embeddings with `sqlite-vec`
You can use a combination of [`vec_slice()`](/api-reference#vec_slice) and
[`vec_normalize()`](/api-reference#vec_slice) on Matryoshka embeddings to
You can use a combination of [`vec_slice()`](../api-reference.md#vec_slice) and
[`vec_normalize()`](../api-reference.md#vec_slice) on Matryoshka embeddings to
truncate.
```sql
select
vec_normalize(vec_slice(title_embeddings, 0, 256)) as title_embeddings_256d
vec_normalize(
vec_slice(title_embeddings, 0, 256)
) as title_embeddings_256d
from vec_articles;
```
[`vec_slice()`](../api-reference.md#vec_slice) will cut down the vector to the first 256 dimensions. Then [`vec_normalize()`](../api-reference.md#vec_normalize) will normalize that truncated vector, which is typically a required step for Matryoshka embeddings.
## Benchmarks
## Suppported Models
@ -47,3 +51,7 @@ https://www.mixedbread.ai/blog/binary-mrl
`mxbai-embed-large-v1`: 1024, 512, 256, 128, 64
`nomic-embed-text-v1.5`: 768, 512, 256, 128, 64
```
# TODO new snowflake model
```