github-mirrors/elasticsearch
1
0
Fork 0
mirror of https://github.com/elastic/elasticsearch.git synced 2025-06-28 17:34:17 -04:00
Code Issues Projects Releases Packages Wiki Activity

New vector_rescore parameter as a quantized index type option (#124581)

Browse source 
This adds a new parameter to the quantized index mapping that allows
default oversampling and rescoring to occur. 

This doesn't adjust any of the defaults. It allows it to be configured.
When the user provides `rescore_vector: {oversample: <number>}` in the
query it will overwrite it.

For example, here is how to use it with bbq:

```
PUT rescored_bbq
{
  "mappings": {
    "properties": {
      "vector": {
        "type": "dense_vector",
        "index_options": {
          "type": "bbq_hnsw",
          "rescore_vector": {"oversample": 3.0}
        }
      }
    }
  }
}
```

Then, when querying, it will auto oversample the `k` by `3x` and rerank
with the raw vectors.

```
POST _search
{
  "knn": {
    "query_vector": [...],
    "field": "vector"
  }
}
```
This commit is contained in:
Benjamin Trent 2025-03-13 09:40:08 -04:00 • committed by GitHub
parent cd25958700
commit b2c1c4e0f0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
13 changed files with 866 additions and 58 deletions
Download patch file Download diff file Expand all files Collapse all files

8
docs/reference/elasticsearch/mapping-reference/dense-vector.md
View file

@ -287,6 +287,14 @@ $$$dense-vector-index-options$$$
`confidence_interval`
: (Optional, float) Only applicable to `int8_hnsw`, `int4_hnsw`, `int8_flat`, and `int4_flat` index types. The confidence interval to use when quantizing the vectors. Can be any value between and including `0.90` and `1.0` or exactly `0`. When the value is `0`, this indicates that dynamic quantiles should be calculated for optimized quantization. When between `0.90` and `1.0`, this value restricts the values used when calculating the quantization thresholds. For example, a value of `0.95` will only use the middle 95% of the values when calculating the quantization thresholds (e.g. the highest and lowest 2.5% of values will be ignored). Defaults to `1/(dims + 1)` for `int8` quantized vectors and `0` for `int4` for dynamic quantile calculation.
`rescore_vector`
: (Optional, object) Functionality in [preview]. An optional section that configures automatic vector rescoring on knn queries for the given field. Only applicable to quantized index types.
:::::{dropdown} Properties of `rescore_vector`
`oversample`
: (required, float) The amount to oversample the search results by. This value should be greater than `1.0` and less than `10.0`. The higher the value, the more vectors will be gathered and rescored with the raw values per shard.
: In case a knn query specifies a `rescore_vector` parameter, the query `rescore_vector` parameter will be used instead.
: See [oversampling and rescoring quantized vectors](docs-content://solutions/search/vector/knn.md#dense-vector-knn-search-rescoring) for details.
:::::
::::