Bump version: 0.17.1-beta.7 → 0.17.1

binary vectors and hamming distance can work on only IVF_FLAT, so
introduce them all in this PR.

---------

Signed-off-by: BubbleCal <bubble-cal@outlook.com>

.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.14.1-beta.4"
+current_version = "0.14.1-beta.6"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

.github/workflows/make-release-commit.yml

@@ -97,3 +97,7 @@ jobs:
         if: ${{ !inputs.dry_run && inputs.other }}
         with:
           github_token: ${{ secrets.GITHUB_TOKEN }}
+      - uses: ./.github/workflows/update_package_lock_nodejs
+        if: ${{ !inputs.dry_run && inputs.other }}
+        with:
+          github_token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/npm-publish.yml

@@ -571,7 +571,7 @@ jobs:
         uses: actions/checkout@v4
         with:
           ref: main
-          persist-credentials: false
+          token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
           fetch-depth: 0
           lfs: true
       - uses: ./.github/workflows/update_package_lock
@@ -589,7 +589,7 @@ jobs:
         uses: actions/checkout@v4
         with:
           ref: main
-          persist-credentials: false
+          token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
           fetch-depth: 0
           lfs: true
       - uses: ./.github/workflows/update_package_lock_nodejs

.github/workflows/rust.yml

@@ -238,3 +238,41 @@ jobs:
           $env:VCPKG_ROOT = $env:VCPKG_INSTALLATION_ROOT
           cargo build --target aarch64-pc-windows-msvc
           cargo test --target aarch64-pc-windows-msvc
+
+  msrv:
+    # Check the minimum supported Rust version
+    name: MSRV Check - Rust v${{ matrix.msrv }}
+    runs-on: ubuntu-24.04
+    strategy:
+      matrix:
+        msrv: ["1.78.0"] # This should match up with rust-version in Cargo.toml
+    env:
+      # Need up-to-date compilers for kernels
+      CC: clang-18
+      CXX: clang++-18
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y protobuf-compiler libssl-dev
+      - name: Install ${{ matrix.msrv }}
+        uses: dtolnay/rust-toolchain@master
+        with:
+          toolchain: ${{ matrix.msrv }}
+      - name: Downgrade  dependencies
+        # These packages have newer requirements for MSRV
+        run: |
+          cargo update -p aws-sdk-bedrockruntime --precise 1.64.0
+          cargo update -p aws-sdk-dynamodb --precise 1.55.0
+          cargo update -p aws-config --precise 1.5.10
+          cargo update -p aws-sdk-kms --precise 1.51.0
+          cargo update -p aws-sdk-s3 --precise 1.65.0
+          cargo update -p aws-sdk-sso --precise 1.50.0
+          cargo update -p aws-sdk-ssooidc --precise 1.51.0
+          cargo update -p aws-sdk-sts --precise 1.51.0
+          cargo update -p home --precise 0.5.9
+      - name: cargo +${{ matrix.msrv }} check
+        run: cargo check --workspace --tests --benches --all-features

Cargo.toml

@@ -18,19 +18,19 @@ repository = "https://github.com/lancedb/lancedb"
 description = "Serverless, low-latency vector database for AI applications"
 keywords = ["lancedb", "lance", "database", "vector", "search"]
 categories = ["database-implementations"]
-rust-version = "1.80.0"                                                     # TODO: lower this once we upgrade Lance again.
+rust-version = "1.78.0"
 
 [workspace.dependencies]
 lance = { "version" = "=0.21.0", "features" = [
     "dynamodb",
-], git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.4" }
+], git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.5" }
-lance-io = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.4" }
+lance-io = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.5" }
-lance-index = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.4" }
+lance-index = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.5" }
-lance-linalg = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.4" }
+lance-linalg = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.5" }
-lance-table = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.4" }
+lance-table = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.5" }
-lance-testing = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.4" }
+lance-testing = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.5" }
-lance-datafusion = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.4" }
+lance-datafusion = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.5" }
-lance-encoding = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.4" }
+lance-encoding = { version = "=0.21.0", git = "https://github.com/lancedb/lance.git", tag = "v0.21.0-beta.5" }
 # Note that this one does not include pyarrow
 arrow = { version = "53.2", optional = false }
 arrow-array = "53.2"

docs/mkdocs.yml

@@ -62,6 +62,7 @@ plugins:
             # for cross references
             - https://arrow.apache.org/docs/objects.inv
             - https://pandas.pydata.org/docs/objects.inv
+            - https://lancedb.github.io/lance/objects.inv
   - mkdocs-jupyter
   - render_swagger:
       allow_arbitrary_locations: true

docs/src/python/python.md

@@ -129,8 +129,12 @@ lists the indices that LanceDb supports.
 
 ::: lancedb.index.LabelList
 
+::: lancedb.index.FTS
+
 ::: lancedb.index.IvfPq
 
+::: lancedb.index.IvfFlat
+
 ## Querying (Asynchronous)
 
 Queries allow you to return data from your database. Basic queries can be

docs/src/python/saas-python.md

@@ -17,4 +17,8 @@ pip install lancedb
 ## Table
 
 ::: lancedb.remote.table.RemoteTable
-
+    options:
+        filters:
+            - "!cleanup_old_versions"
+            - "!compact_files"
+            - "!optimize"

docs/src/search.md

@@ -14,10 +14,14 @@ Distance metrics are a measure of the similarity between a pair of vectors.
 Currently, LanceDB supports the following metrics:
 
 | Metric    | Description                                                                 |
-| -------- | --------------------------------------------------------------------------- |
+| --------- | --------------------------------------------------------------------------- |
 | `l2`      | [Euclidean / L2 distance](https://en.wikipedia.org/wiki/Euclidean_distance) |
 | `cosine`  | [Cosine Similarity](https://en.wikipedia.org/wiki/Cosine_similarity)        |
 | `dot`     | [Dot Production](https://en.wikipedia.org/wiki/Dot_product)                 |
+| `hamming` | [Hamming Distance](https://en.wikipedia.org/wiki/Hamming_distance)          |
+
+!!! note
+    The `hamming` metric is only available for binary vectors.
 
 ## Exhaustive search (kNN)
 
@@ -107,6 +111,31 @@ an ANN search means that using an index often involves a trade-off between recal
 See the [IVF_PQ index](./concepts/index_ivfpq.md) for a deeper description of how `IVF_PQ`
 indexes work in LanceDB.
 
+## Binary vector
+
+LanceDB supports binary vectors as a data type, and has the ability to search binary vectors with hamming distance. The binary vectors are stored as uint8 arrays (every 8 bits are stored as a byte):
+
+!!! note
+    The dim of the binary vector must be a multiple of 8. A vector of dim 128 will be stored as a uint8 array of size 16.
+
+=== "Python"
+
+    === "sync API"
+
+        ```python
+        --8<-- "python/python/tests/docs/test_binary_vector.py:imports"
+
+        --8<-- "python/python/tests/docs/test_binary_vector.py:sync_binary_vector"
+        ```
+
+    === "async API"
+
+        ```python
+        --8<-- "python/python/tests/docs/test_binary_vector.py:imports"
+
+        --8<-- "python/python/tests/docs/test_binary_vector.py:async_binary_vector"
+        ```
+
 ## Output search results
 
 LanceDB returns vector search results via different formats commonly used in python.

docs/test/md_testing.py

@@ -16,6 +16,7 @@ excluded_globs = [
     "../src/concepts/*.md",
     "../src/ann_indexes.md",
     "../src/basic.md",
+    "../src/search.md",
     "../src/hybrid_search/hybrid_search.md",
     "../src/reranking/*.md",
     "../src/guides/tuning_retrievers/*.md",

java/core/pom.xml

@@ -8,7 +8,7 @@
     <parent>
         <groupId>com.lancedb</groupId>
         <artifactId>lancedb-parent</artifactId>
-        <version>0.14.1-beta.4</version>
+        <version>0.14.1-beta.6</version>
         <relativePath>../pom.xml</relativePath>
     </parent>
 

java/pom.xml

@@ -6,7 +6,7 @@
 
     <groupId>com.lancedb</groupId>
     <artifactId>lancedb-parent</artifactId>
-    <version>0.14.1-beta.4</version>
+    <version>0.14.1-beta.6</version>
     <packaging>pom</packaging>
 
     <name>LanceDB Parent</name>

node/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "vectordb",
-  "version": "0.14.1-beta.4",
+  "version": "0.14.1-beta.6",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "vectordb",
-      "version": "0.14.1-beta.4",
+      "version": "0.14.1-beta.6",
       "cpu": [
         "x64",
         "arm64"
@@ -52,14 +52,14 @@
         "uuid": "^9.0.0"
       },
       "optionalDependencies": {
-        "@lancedb/vectordb-darwin-arm64": "0.14.1-beta.4",
+        "@lancedb/vectordb-darwin-arm64": "0.14.1-beta.6",
-        "@lancedb/vectordb-darwin-x64": "0.14.1-beta.4",
+        "@lancedb/vectordb-darwin-x64": "0.14.1-beta.6",
-        "@lancedb/vectordb-linux-arm64-gnu": "0.14.1-beta.4",
+        "@lancedb/vectordb-linux-arm64-gnu": "0.14.1-beta.6",
-        "@lancedb/vectordb-linux-arm64-musl": "0.14.1-beta.4",
+        "@lancedb/vectordb-linux-arm64-musl": "0.14.1-beta.6",
-        "@lancedb/vectordb-linux-x64-gnu": "0.14.1-beta.4",
+        "@lancedb/vectordb-linux-x64-gnu": "0.14.1-beta.6",
-        "@lancedb/vectordb-linux-x64-musl": "0.14.1-beta.4",
+        "@lancedb/vectordb-linux-x64-musl": "0.14.1-beta.6",
-        "@lancedb/vectordb-win32-arm64-msvc": "0.14.1-beta.4",
+        "@lancedb/vectordb-win32-arm64-msvc": "0.14.1-beta.6",
-        "@lancedb/vectordb-win32-x64-msvc": "0.14.1-beta.4"
+        "@lancedb/vectordb-win32-x64-msvc": "0.14.1-beta.6"
       },
       "peerDependencies": {
         "@apache-arrow/ts": "^14.0.2",

node/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vectordb",
-  "version": "0.14.1-beta.4",
+  "version": "0.14.1-beta.6",
   "description": " Serverless, low-latency vector database for AI applications",
   "private": false,
   "main": "dist/index.js",
@@ -92,13 +92,13 @@
     }
   },
   "optionalDependencies": {
-    "@lancedb/vectordb-darwin-x64": "0.14.1-beta.4",
+    "@lancedb/vectordb-darwin-x64": "0.14.1-beta.6",
-    "@lancedb/vectordb-darwin-arm64": "0.14.1-beta.4",
+    "@lancedb/vectordb-darwin-arm64": "0.14.1-beta.6",
-    "@lancedb/vectordb-linux-x64-gnu": "0.14.1-beta.4",
+    "@lancedb/vectordb-linux-x64-gnu": "0.14.1-beta.6",
-    "@lancedb/vectordb-linux-arm64-gnu": "0.14.1-beta.4",
+    "@lancedb/vectordb-linux-arm64-gnu": "0.14.1-beta.6",
-    "@lancedb/vectordb-linux-x64-musl": "0.14.1-beta.4",
+    "@lancedb/vectordb-linux-x64-musl": "0.14.1-beta.6",
-    "@lancedb/vectordb-linux-arm64-musl": "0.14.1-beta.4",
+    "@lancedb/vectordb-linux-arm64-musl": "0.14.1-beta.6",
-    "@lancedb/vectordb-win32-x64-msvc": "0.14.1-beta.4",
+    "@lancedb/vectordb-win32-x64-msvc": "0.14.1-beta.6",
-    "@lancedb/vectordb-win32-arm64-msvc": "0.14.1-beta.4"
+    "@lancedb/vectordb-win32-arm64-msvc": "0.14.1-beta.6"
   }
 }

nodejs/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "lancedb-nodejs"
 edition.workspace = true
-version = "0.14.1-beta.4"
+version = "0.14.1-beta.6"
 license.workspace = true
 description.workspace = true
 repository.workspace = true

nodejs/npm/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.14.1-beta.4",
+	"version": "0.14.1-beta.6",
 	"os": ["darwin"],
 	"cpu": ["arm64"],
 	"main": "lancedb.darwin-arm64.node",

nodejs/npm/darwin-x64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-x64",
-	"version": "0.14.1-beta.4",
+	"version": "0.14.1-beta.6",
 	"os": ["darwin"],
 	"cpu": ["x64"],
 	"main": "lancedb.darwin-x64.node",

nodejs/npm/linux-arm64-gnu/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-gnu",
-	"version": "0.14.1-beta.4",
+	"version": "0.14.1-beta.6",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-gnu.node",

nodejs/npm/linux-arm64-musl/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-musl",
-	"version": "0.14.1-beta.4",
+	"version": "0.14.1-beta.6",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-musl.node",

nodejs/npm/linux-x64-gnu/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-gnu",
-	"version": "0.14.1-beta.4",
+	"version": "0.14.1-beta.6",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-gnu.node",

nodejs/npm/linux-x64-musl/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-musl",
-	"version": "0.14.1-beta.4",
+	"version": "0.14.1-beta.6",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-musl.node",

nodejs/npm/win32-arm64-msvc/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lancedb/lancedb-win32-arm64-msvc",
-  "version": "0.14.1-beta.4",
+  "version": "0.14.1-beta.6",
   "os": [
     "win32"
   ],

nodejs/npm/win32-x64-msvc/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.14.1-beta.4",
+	"version": "0.14.1-beta.6",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",

nodejs/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@lancedb/lancedb",
-  "version": "0.14.0",
+  "version": "0.14.1-beta.6",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@lancedb/lancedb",
-      "version": "0.14.0",
+      "version": "0.14.1-beta.6",
       "cpu": [
         "x64",
         "arm64"

nodejs/package.json

@@ -11,7 +11,7 @@
     "ann"
   ],
   "private": false,
-  "version": "0.14.1-beta.4",
+  "version": "0.14.1-beta.6",
   "main": "dist/index.js",
   "exports": {
     ".": "./dist/index.js",

nodejs/src/util.rs

@@ -5,8 +5,9 @@ pub fn parse_distance_type(distance_type: impl AsRef<str>) -> napi::Result<Dista
         "l2" => Ok(DistanceType::L2),
         "cosine" => Ok(DistanceType::Cosine),
         "dot" => Ok(DistanceType::Dot),
+        "hamming" => Ok(DistanceType::Hamming),
         _ => Err(napi::Error::from_reason(format!(
-            "Invalid distance type '{}'.  Must be one of l2, cosine, or dot",
+            "Invalid distance type '{}'.  Must be one of l2, cosine, dot, or hamming",
             distance_type.as_ref()
         ))),
     }

python/.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.17.1-beta.5"
+current_version = "0.17.1"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

python/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.17.1-beta.5"
+version = "0.17.1"
 edition.workspace = true
 description = "Python bindings for LanceDB"
 license.workspace = true

python/pyproject.toml

@@ -3,7 +3,7 @@ name = "lancedb"
 # version in Cargo.toml
 dependencies = [
     "deprecation",
-    "pylance==0.21.0b4",
+    "pylance==0.21.0b5",
     "tqdm>=4.27.0",
     "pydantic>=1.10",
     "packaging",

python/python/lancedb/db.py

@@ -18,12 +18,12 @@ from pathlib import Path
 from typing import TYPE_CHECKING, Dict, Iterable, List, Literal, Optional, Union
 
 from lancedb.embeddings.registry import EmbeddingFunctionRegistry
-from overrides import EnforceOverrides, override
+from overrides import EnforceOverrides, override  # type: ignore
 
 from lancedb.common import data_to_reader, sanitize_uri, validate_schema
 from lancedb.background_loop import LOOP
 
-from ._lancedb import connect as lancedb_connect
+from ._lancedb import connect as lancedb_connect  # type: ignore
 from .table import (
     AsyncTable,
     LanceTable,
@@ -503,13 +503,7 @@ class LanceDBConnection(DBConnection):
         ignore_missing: bool, default False
             If True, ignore if the table does not exist.
         """
-        try:
+        LOOP.run(self._conn.drop_table(name, ignore_missing=ignore_missing))
-            LOOP.run(self._conn.drop_table(name))
-        except ValueError as e:
-            if not ignore_missing:
-                raise e
-            if f"Table '{name}' was not found" not in str(e):
-                raise e
 
     @override
     def drop_database(self):
@@ -886,15 +880,23 @@ class AsyncConnection(object):
         """
         await self._inner.rename_table(old_name, new_name)
 
-    async def drop_table(self, name: str):
+    async def drop_table(self, name: str, *, ignore_missing: bool = False):
         """Drop a table from the database.
 
         Parameters
         ----------
         name: str
             The name of the table.
+        ignore_missing: bool, default False
+            If True, ignore if the table does not exist.
         """
+        try:
             await self._inner.drop_table(name)
+        except ValueError as e:
+            if not ignore_missing:
+                raise e
+            if f"Table '{name}' was not found" not in str(e):
+                raise e
 
     async def drop_database(self):
         """

python/python/lancedb/index.py

@@ -355,6 +355,97 @@ class HnswSq:
     ef_construction: int = 300
 
 
+@dataclass
+class IvfFlat:
+    """Describes an IVF Flat Index
+
+    This index stores raw vectors.
+    These vectors are grouped into partitions of similar vectors.
+    Each partition keeps track of a centroid which is
+    the average value of all vectors in the group.
+
+    Attributes
+    ----------
+    distance_type: str, default "L2"
+        The distance metric used to train the index
+
+        This is used when training the index to calculate the IVF partitions
+        (vectors are grouped in partitions with similar vectors according to this
+        distance type) and to calculate a subvector's code during quantization.
+
+        The distance type used to train an index MUST match the distance type used
+        to search the index.  Failure to do so will yield inaccurate results.
+
+        The following distance types are available:
+
+        "l2" - Euclidean distance. This is a very common distance metric that
+        accounts for both magnitude and direction when determining the distance
+        between vectors. L2 distance has a range of [0, ∞).
+
+        "cosine" - Cosine distance.  Cosine distance is a distance metric
+        calculated from the cosine similarity between two vectors. Cosine
+        similarity is a measure of similarity between two non-zero vectors of an
+        inner product space. It is defined to equal the cosine of the angle
+        between them.  Unlike L2, the cosine distance is not affected by the
+        magnitude of the vectors.  Cosine distance has a range of [0, 2].
+
+        Note: the cosine distance is undefined when one (or both) of the vectors
+        are all zeros (there is no direction).  These vectors are invalid and may
+        never be returned from a vector search.
+
+        "dot" - Dot product. Dot distance is the dot product of two vectors. Dot
+        distance has a range of (-∞, ∞). If the vectors are normalized (i.e. their
+        L2 norm is 1), then dot distance is equivalent to the cosine distance.
+
+        "hamming" - Hamming distance. Hamming distance is a distance metric
+        calculated as the number of positions at which the corresponding bits are
+        different. Hamming distance has a range of [0, vector dimension].
+
+    num_partitions: int, default sqrt(num_rows)
+        The number of IVF partitions to create.
+
+        This value should generally scale with the number of rows in the dataset.
+        By default the number of partitions is the square root of the number of
+        rows.
+
+        If this value is too large then the first part of the search (picking the
+        right partition) will be slow.  If this value is too small then the second
+        part of the search (searching within a partition) will be slow.
+
+    max_iterations: int, default 50
+        Max iteration to train kmeans.
+
+        When training an IVF PQ index we use kmeans to calculate the partitions.
+        This parameter controls how many iterations of kmeans to run.
+
+        Increasing this might improve the quality of the index but in most cases
+        these extra iterations have diminishing returns.
+
+        The default value is 50.
+    sample_rate: int, default 256
+        The rate used to calculate the number of training vectors for kmeans.
+
+        When an IVF PQ index is trained, we need to calculate partitions.  These
+        are groups of vectors that are similar to each other.  To do this we use an
+        algorithm called kmeans.
+
+        Running kmeans on a large dataset can be slow.  To speed this up we run
+        kmeans on a random sample of the data.  This parameter controls the size of
+        the sample.  The total number of vectors used to train the index is
+        `sample_rate * num_partitions`.
+
+        Increasing this value might improve the quality of the index but in most
+        cases the default should be sufficient.
+
+        The default value is 256.
+    """
+
+    distance_type: Literal["l2", "cosine", "dot", "hamming"] = "l2"
+    num_partitions: Optional[int] = None
+    max_iterations: int = 50
+    sample_rate: int = 256
+
+
 @dataclass
 class IvfPq:
     """Describes an IVF PQ Index
@@ -477,4 +568,4 @@ class IvfPq:
     sample_rate: int = 256
 
 
-__all__ = ["BTree", "IvfPq", "HnswPq", "HnswSq", "IndexConfig"]
+__all__ = ["BTree", "IvfFlat", "IvfPq", "HnswPq", "HnswSq", "IndexConfig"]

python/python/lancedb/query.py

@@ -126,6 +126,9 @@ class Query(pydantic.BaseModel):
 
     ef: Optional[int] = None
 
+    # Default is true. Set to false to enforce a brute force search.
+    use_index: bool = True
+
 
 class LanceQueryBuilder(ABC):
     """An abstract query builder. Subclasses are defined for vector search,
@@ -253,6 +256,7 @@ class LanceQueryBuilder(ABC):
         self._vector = None
         self._text = None
         self._ef = None
+        self._use_index = True
 
     @deprecation.deprecated(
         deprecated_in="0.3.1",
@@ -511,6 +515,7 @@ class LanceQueryBuilder(ABC):
                 "metric": self._metric,
                 "nprobes": self._nprobes,
                 "refine_factor": self._refine_factor,
+                "use_index": self._use_index,
             },
             prefilter=self._prefilter,
             filter=self._str_query,
@@ -729,6 +734,7 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
             offset=self._offset,
             fast_search=self._fast_search,
             ef=self._ef,
+            use_index=self._use_index,
         )
         result_set = self._table._execute_query(query, batch_size)
         if self._reranker is not None:
@@ -802,6 +808,24 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         self._str_query = query_string if query_string is not None else self._str_query
         return self
 
+    def bypass_vector_index(self) -> LanceVectorQueryBuilder:
+        """
+        If this is called then any vector index is skipped
+
+        An exhaustive (flat) search will be performed.  The query vector will
+        be compared to every vector in the table.  At high scales this can be
+        expensive.  However, this is often still useful.  For example, skipping
+        the vector index can give you ground truth results which you can use to
+        calculate your recall to select an appropriate value for nprobes.
+
+        Returns
+        -------
+        LanceVectorQueryBuilder
+            The LanceVectorQueryBuilder object.
+        """
+        self._use_index = False
+        return self
+
 
 class LanceFtsQueryBuilder(LanceQueryBuilder):
     """A builder for full text search for LanceDB."""
@@ -1108,6 +1132,8 @@ class LanceHybridQueryBuilder(LanceQueryBuilder):
             self._vector_query.refine_factor(self._refine_factor)
         if self._ef:
             self._vector_query.ef(self._ef)
+        if not self._use_index:
+            self._vector_query.bypass_vector_index()
 
         with ThreadPoolExecutor() as executor:
             fts_future = executor.submit(self._fts_query.with_row_id(True).to_arrow)
@@ -1323,6 +1349,24 @@ class LanceHybridQueryBuilder(LanceQueryBuilder):
         self._text = text
         return self
 
+    def bypass_vector_index(self) -> LanceHybridQueryBuilder:
+        """
+        If this is called then any vector index is skipped
+
+        An exhaustive (flat) search will be performed.  The query vector will
+        be compared to every vector in the table.  At high scales this can be
+        expensive.  However, this is often still useful.  For example, skipping
+        the vector index can give you ground truth results which you can use to
+        calculate your recall to select an appropriate value for nprobes.
+
+        Returns
+        -------
+        LanceHybridQueryBuilder
+            The LanceHybridQueryBuilder object.
+        """
+        self._use_index = False
+        return self
+
 
 class AsyncQueryBase(object):
     def __init__(self, inner: Union[LanceQuery | LanceVectorQuery]):

python/python/lancedb/remote/table.py

@@ -15,6 +15,7 @@ from datetime import timedelta
 import logging
 from functools import cached_property
 from typing import Dict, Iterable, List, Optional, Union, Literal
+import warnings
 
 from lancedb._lancedb import IndexConfig
 from lancedb.embeddings.base import EmbeddingFunctionConfig
@@ -481,16 +482,28 @@ class RemoteTable(Table):
         )
 
     def cleanup_old_versions(self, *_):
-        """cleanup_old_versions() is not supported on the LanceDB cloud"""
+        """
-        raise NotImplementedError(
+        cleanup_old_versions() is a no-op on LanceDB Cloud.
-            "cleanup_old_versions() is not supported on the LanceDB cloud"
+
+        Tables are automatically cleaned up and optimized.
+        """
+        warnings.warn(
+            "cleanup_old_versions() is a no-op on LanceDB Cloud. "
+            "Tables are automatically cleaned up and optimized."
         )
+        pass
 
     def compact_files(self, *_):
-        """compact_files() is not supported on the LanceDB cloud"""
+        """
-        raise NotImplementedError(
+        compact_files() is a no-op on LanceDB Cloud.
-            "compact_files() is not supported on the LanceDB cloud"
+
+        Tables are automatically compacted and optimized.
+        """
+        warnings.warn(
+            "compact_files() is a no-op on LanceDB Cloud. "
+            "Tables are automatically compacted and optimized."
         )
+        pass
 
     def optimize(
         self,
@@ -498,12 +511,16 @@ class RemoteTable(Table):
         cleanup_older_than: Optional[timedelta] = None,
         delete_unverified: bool = False,
     ):
-        """optimize() is not supported on the LanceDB cloud.
+        """
-        Indices are optimized automatically."""
+        optimize() is a no-op on LanceDB Cloud.
-        raise NotImplementedError(
+
-            "optimize() is not supported on the LanceDB cloud. "
+        Indices are optimized automatically.
+        """
+        warnings.warn(
+            "optimize() is a no-op on LanceDB Cloud. "
             "Indices are optimized automatically."
         )
+        pass
 
     def count_rows(self, filter: Optional[str] = None) -> int:
         return LOOP.run(self._table.count_rows(filter))

python/python/lancedb/table.py

@@ -34,7 +34,7 @@ from lance.dependencies import _check_for_hugging_face
 
 from .common import DATA, VEC, VECTOR_COLUMN_NAME
 from .embeddings import EmbeddingFunctionConfig, EmbeddingFunctionRegistry
-from .index import BTree, IvfPq, Bitmap, LabelList, HnswPq, HnswSq, FTS
+from .index import BTree, IvfFlat, IvfPq, Bitmap, LabelList, HnswPq, HnswSq, FTS
 from .merge import LanceMergeInsertBuilder
 from .pydantic import LanceModel, model_to_dict
 from .query import (
@@ -433,7 +433,9 @@ class Table(ABC):
         accelerator: Optional[str] = None,
         index_cache_size: Optional[int] = None,
         *,
-        index_type: Literal["IVF_PQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ"] = "IVF_PQ",
+        index_type: Literal[
+            "IVF_FLAT", "IVF_PQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ"
+        ] = "IVF_PQ",
         num_bits: int = 8,
         max_iterations: int = 50,
         sample_rate: int = 256,
@@ -446,8 +448,9 @@ class Table(ABC):
         ----------
         metric: str, default "L2"
             The distance metric to use when creating the index.
-            Valid values are "L2", "cosine", or "dot".
+            Valid values are "L2", "cosine", "dot", or "hamming".
             L2 is euclidean distance.
+            Hamming is available only for binary vectors.
         num_partitions: int, default 256
             The number of IVF partitions to use when creating the index.
             Default is 256.
@@ -917,9 +920,6 @@ class Table(ABC):
         """
         Clean up old versions of the table, freeing disk space.
 
-        Note: This function is not available in LanceDb Cloud (since LanceDb
-        Cloud manages cleanup for you automatically)
-
         Parameters
         ----------
         older_than: timedelta, default None
@@ -936,21 +936,38 @@ class Table(ABC):
         CleanupStats
             The stats of the cleanup operation, including how many bytes were
             freed.
+
+        See Also
+        --------
+        [Table.optimize][lancedb.table.Table.optimize]: A more comprehensive
+            optimization operation that includes cleanup as well as other operations.
+
+        Notes
+        -----
+        This function is not available in LanceDb Cloud (since LanceDB
+        Cloud manages cleanup for you automatically)
         """
 
     @abstractmethod
     def compact_files(self, *args, **kwargs):
         """
         Run the compaction process on the table.
-
-        Note: This function is not available in LanceDb Cloud (since LanceDb
-        Cloud manages compaction for you automatically)
-
         This can be run after making several small appends to optimize the table
         for faster reads.
 
-        Arguments are passed onto :meth:`lance.dataset.DatasetOptimizer.compact_files`.
+        Arguments are passed onto Lance's
+        [compact_files][lance.dataset.DatasetOptimizer.compact_files].
         For most cases, the default should be fine.
+
+        See Also
+        --------
+        [Table.optimize][lancedb.table.Table.optimize]: A more comprehensive
+            optimization operation that includes cleanup as well as other operations.
+
+        Notes
+        -----
+        This function is not available in LanceDB Cloud (since LanceDB
+        Cloud manages compaction for you automatically)
         """
 
     @abstractmethod
@@ -1394,7 +1411,9 @@ class LanceTable(Table):
         accelerator: Optional[str] = None,
         index_cache_size: Optional[int] = None,
         num_bits: int = 8,
-        index_type: Literal["IVF_PQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ"] = "IVF_PQ",
+        index_type: Literal[
+            "IVF_FLAT", "IVF_PQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ"
+        ] = "IVF_PQ",
         max_iterations: int = 50,
         sample_rate: int = 256,
         m: int = 20,
@@ -1418,6 +1437,13 @@ class LanceTable(Table):
             )
             self.checkout_latest()
             return
+        elif index_type == "IVF_FLAT":
+            config = IvfFlat(
+                distance_type=metric,
+                num_partitions=num_partitions,
+                max_iterations=max_iterations,
+                sample_rate=sample_rate,
+            )
         elif index_type == "IVF_PQ":
             config = IvfPq(
                 distance_type=metric,
@@ -2605,7 +2631,7 @@ class AsyncTable:
         *,
         replace: Optional[bool] = None,
         config: Optional[
-            Union[IvfPq, HnswPq, HnswSq, BTree, Bitmap, LabelList, FTS]
+            Union[IvfFlat, IvfPq, HnswPq, HnswSq, BTree, Bitmap, LabelList, FTS]
         ] = None,
     ):
         """Create an index to speed up queries
@@ -2634,7 +2660,7 @@ class AsyncTable:
         """
         if config is not None:
             if not isinstance(
-                config, (IvfPq, HnswPq, HnswSq, BTree, Bitmap, LabelList, FTS)
+                config, (IvfFlat, IvfPq, HnswPq, HnswSq, BTree, Bitmap, LabelList, FTS)
             ):
                 raise TypeError(
                     "config must be an instance of IvfPq, HnswPq, HnswSq, BTree,"
@@ -2798,6 +2824,8 @@ class AsyncTable:
                 async_query = async_query.column(query.vector_column)
             if query.ef:
                 async_query = async_query.ef(query.ef)
+            if not query.use_index:
+                async_query = async_query.bypass_vector_index()
 
         if not query.prefilter:
             async_query = async_query.postfilter()

python/python/tests/docs/test_binary_vector.py

@@ -0,0 +1,44 @@
+import shutil
+
+# --8<-- [start:imports]
+import lancedb
+import numpy as np
+import pytest
+# --8<-- [end:imports]
+
+shutil.rmtree("data/binary_lancedb", ignore_errors=True)
+
+
+def test_binary_vector():
+    # --8<-- [start:sync_binary_vector]
+    db = lancedb.connect("data/binary_lancedb")
+    data = [
+        {
+            "id": i,
+            "vector": np.random.randint(0, 256, size=16),
+        }
+        for i in range(1024)
+    ]
+    tbl = db.create_table("my_binary_vectors", data=data)
+    query = np.random.randint(0, 256, size=16)
+    tbl.search(query).to_arrow()
+    # --8<-- [end:sync_binary_vector]
+    db.drop_table("my_binary_vectors")
+
+
+@pytest.mark.asyncio
+async def test_binary_vector_async():
+    # --8<-- [start:async_binary_vector]
+    db = await lancedb.connect_async("data/binary_lancedb")
+    data = [
+        {
+            "id": i,
+            "vector": np.random.randint(0, 256, size=16),
+        }
+        for i in range(1024)
+    ]
+    tbl = await db.create_table("my_binary_vectors", data=data)
+    query = np.random.randint(0, 256, size=16)
+    await tbl.query().nearest_to(query).to_arrow()
+    # --8<-- [end:async_binary_vector]
+    await db.drop_table("my_binary_vectors")

python/python/tests/test_db.py

@@ -508,6 +508,32 @@ def test_delete_table(tmp_db: lancedb.DBConnection):
     tmp_db.drop_table("does_not_exist", ignore_missing=True)
 
 
+@pytest.mark.asyncio
+async def test_delete_table_async(tmp_db: lancedb.DBConnection):
+    data = pd.DataFrame(
+        {
+            "vector": [[3.1, 4.1], [5.9, 26.5]],
+            "item": ["foo", "bar"],
+            "price": [10.0, 20.0],
+        }
+    )
+
+    tmp_db.create_table("test", data=data)
+
+    with pytest.raises(Exception):
+        tmp_db.create_table("test", data=data)
+
+    assert tmp_db.table_names() == ["test"]
+
+    tmp_db.drop_table("test")
+    assert tmp_db.table_names() == []
+
+    tmp_db.create_table("test", data=data)
+    assert tmp_db.table_names() == ["test"]
+
+    tmp_db.drop_table("does_not_exist", ignore_missing=True)
+
+
 def test_drop_database(tmp_db: lancedb.DBConnection):
     data = pd.DataFrame(
         {
@@ -681,3 +707,25 @@ def test_create_table_with_invalid_names(tmp_db: lancedb.DBConnection):
     with pytest.raises(ValueError):
         tmp_db.create_table("foo$$bar", data)
     tmp_db.create_table("foo.bar", data)
+
+
+def test_bypass_vector_index_sync(tmp_db: lancedb.DBConnection):
+    data = [{"vector": np.random.rand(32)} for _ in range(512)]
+    sample_key = data[100]["vector"]
+    table = tmp_db.create_table(
+        "test",
+        data,
+    )
+
+    table.create_index(
+        num_partitions=2,
+        num_sub_vectors=2,
+    )
+
+    plan_with_index = table.search(sample_key).explain_plan(verbose=True)
+    assert "ANN" in plan_with_index
+
+    plan_without_index = (
+        table.search(sample_key).bypass_vector_index().explain_plan(verbose=True)
+    )
+    assert "KNN" in plan_without_index

python/python/tests/test_index.py

@@ -8,7 +8,7 @@ import pyarrow as pa
 import pytest
 import pytest_asyncio
 from lancedb import AsyncConnection, AsyncTable, connect_async
-from lancedb.index import BTree, IvfPq, Bitmap, LabelList, HnswPq, HnswSq
+from lancedb.index import BTree, IvfFlat, IvfPq, Bitmap, LabelList, HnswPq, HnswSq
 
 
 @pytest_asyncio.fixture
@@ -42,6 +42,27 @@ async def some_table(db_async):
     )
 
 
+@pytest_asyncio.fixture
+async def binary_table(db_async):
+    data = [
+        {
+            "id": i,
+            "vector": [i] * 128,
+        }
+        for i in range(NROWS)
+    ]
+    return await db_async.create_table(
+        "binary_table",
+        data,
+        schema=pa.schema(
+            [
+                pa.field("id", pa.int64()),
+                pa.field("vector", pa.list_(pa.uint8(), 128)),
+            ]
+        ),
+    )
+
+
 @pytest.mark.asyncio
 async def test_create_scalar_index(some_table: AsyncTable):
     # Can create
@@ -143,3 +164,27 @@ async def test_create_hnswsq_index(some_table: AsyncTable):
     await some_table.create_index("vector", config=HnswSq(num_partitions=10))
     indices = await some_table.list_indices()
     assert len(indices) == 1
+
+
+@pytest.mark.asyncio
+async def test_create_index_with_binary_vectors(binary_table: AsyncTable):
+    await binary_table.create_index(
+        "vector", config=IvfFlat(distance_type="hamming", num_partitions=10)
+    )
+    indices = await binary_table.list_indices()
+    assert len(indices) == 1
+    assert indices[0].index_type == "IvfFlat"
+    assert indices[0].columns == ["vector"]
+    assert indices[0].name == "vector_idx"
+
+    stats = await binary_table.index_stats("vector_idx")
+    assert stats.index_type == "IVF_FLAT"
+    assert stats.distance_type == "hamming"
+    assert stats.num_indexed_rows == await binary_table.count_rows()
+    assert stats.num_unindexed_rows == 0
+    assert stats.num_indices == 1
+
+    # the dataset contains vectors with all values from 0 to 255
+    for v in range(256):
+        res = await binary_table.query().nearest_to([v] * 128).to_arrow()
+        assert res["id"][0].as_py() == v

python/src/index.rs

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use lancedb::index::vector::IvfFlatIndexBuilder;
 use lancedb::index::{
     scalar::{BTreeIndexBuilder, FtsIndexBuilder, TokenizerConfig},
     vector::{IvfHnswPqIndexBuilder, IvfHnswSqIndexBuilder, IvfPqIndexBuilder},
@@ -59,6 +60,18 @@ pub fn extract_index_params(source: &Option<Bound<'_, PyAny>>) -> PyResult<Lance
                 opts.tokenizer_configs = inner_opts;
                 Ok(LanceDbIndex::FTS(opts))
             },
+            "IvfFlat" => {
+                let params = source.extract::<IvfFlatParams>()?;
+                let distance_type = parse_distance_type(params.distance_type)?;
+                let mut ivf_flat_builder = IvfFlatIndexBuilder::default()
+                    .distance_type(distance_type)
+                    .max_iterations(params.max_iterations)
+                    .sample_rate(params.sample_rate);
+                if let Some(num_partitions) = params.num_partitions {
+                    ivf_flat_builder = ivf_flat_builder.num_partitions(num_partitions);
+                }
+                Ok(LanceDbIndex::IvfFlat(ivf_flat_builder))
+            },
             "IvfPq" => {
                 let params = source.extract::<IvfPqParams>()?;
                 let distance_type = parse_distance_type(params.distance_type)?;
@@ -129,6 +142,14 @@ struct FtsParams {
     ascii_folding: bool,
 }
 
+#[derive(FromPyObject)]
+struct IvfFlatParams {
+    distance_type: String,
+    num_partitions: Option<u32>,
+    max_iterations: u32,
+    sample_rate: u32,
+}
+
 #[derive(FromPyObject)]
 struct IvfPqParams {
     distance_type: String,

python/src/util.rs

@@ -43,8 +43,9 @@ pub fn parse_distance_type(distance_type: impl AsRef<str>) -> PyResult<DistanceT
         "l2" => Ok(DistanceType::L2),
         "cosine" => Ok(DistanceType::Cosine),
         "dot" => Ok(DistanceType::Dot),
+        "hamming" => Ok(DistanceType::Hamming),
         _ => Err(PyValueError::new_err(format!(
-            "Invalid distance type '{}'.  Must be one of l2, cosine, or dot",
+            "Invalid distance type '{}'.  Must be one of l2, cosine, dot, or hamming",
             distance_type.as_ref()
         ))),
     }

rust-toolchain.toml

@@ -1,2 +1,2 @@
 [toolchain]
-channel = "1.80.0"
+channel = "1.83.0"

rust/ffi/node/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-node"
-version = "0.14.1-beta.4"
+version = "0.14.1-beta.6"
 description = "Serverless, low-latency vector database for AI applications"
 license.workspace = true
 edition.workspace = true

rust/lancedb/Cargo.toml

@@ -1,13 +1,13 @@
 [package]
 name = "lancedb"
-version = "0.14.1-beta.4"
+version = "0.14.1-beta.6"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true
 repository.workspace = true
 keywords.workspace = true
 categories.workspace = true
-rust-version = "1.75"
+rust-version.workspace = true
 
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 [dependencies]

rust/lancedb/src/index.rs

@@ -17,6 +17,7 @@ use std::sync::Arc;
 use scalar::FtsIndexBuilder;
 use serde::Deserialize;
 use serde_with::skip_serializing_none;
+use vector::IvfFlatIndexBuilder;
 
 use crate::{table::TableInternal, DistanceType, Error, Result};
 
@@ -56,6 +57,9 @@ pub enum Index {
     /// Full text search index using bm25.
     FTS(FtsIndexBuilder),
 
+    /// IVF index
+    IvfFlat(IvfFlatIndexBuilder),
+
     /// IVF index with Product Quantization
     IvfPq(IvfPqIndexBuilder),
 
@@ -106,6 +110,8 @@ impl IndexBuilder {
 #[derive(Debug, Clone, PartialEq, Deserialize)]
 pub enum IndexType {
     // Vector
+    #[serde(alias = "IVF_FLAT")]
+    IvfFlat,
     #[serde(alias = "IVF_PQ")]
     IvfPq,
     #[serde(alias = "IVF_HNSW_PQ")]
@@ -127,6 +133,7 @@ pub enum IndexType {
 impl std::fmt::Display for IndexType {
     fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
         match self {
+            Self::IvfFlat => write!(f, "IVF_FLAT"),
             Self::IvfPq => write!(f, "IVF_PQ"),
             Self::IvfHnswPq => write!(f, "IVF_HNSW_PQ"),
             Self::IvfHnswSq => write!(f, "IVF_HNSW_SQ"),
@@ -147,6 +154,7 @@ impl std::str::FromStr for IndexType {
             "BITMAP" => Ok(Self::Bitmap),
             "LABEL_LIST" | "LABELLIST" => Ok(Self::LabelList),
             "FTS" | "INVERTED" => Ok(Self::FTS),
+            "IVF_FLAT" => Ok(Self::IvfFlat),
             "IVF_PQ" => Ok(Self::IvfPq),
             "IVF_HNSW_PQ" => Ok(Self::IvfHnswPq),
             "IVF_HNSW_SQ" => Ok(Self::IvfHnswSq),

rust/lancedb/src/index/vector.rs

@@ -162,6 +162,43 @@ macro_rules! impl_hnsw_params_setter {
     };
 }
 
+/// Builder for an IVF Flat index.
+///
+/// This index stores raw vectors. These vectors are grouped into partitions of similar vectors.
+/// Each partition keeps track of a centroid which is the average value of all vectors in the group.
+///
+/// During a query the centroids are compared with the query vector to find the closest partitions.
+/// The raw vectors in these partitions are then searched to find the closest vectors.
+///
+/// The partitioning process is called IVF and the `num_partitions` parameter controls how many groups to create.
+///
+/// Note that training an IVF Flat index on a large dataset is a slow operation and currently is also a memory intensive operation.
+#[derive(Debug, Clone)]
+pub struct IvfFlatIndexBuilder {
+    pub(crate) distance_type: DistanceType,
+
+    // IVF
+    pub(crate) num_partitions: Option<u32>,
+    pub(crate) sample_rate: u32,
+    pub(crate) max_iterations: u32,
+}
+
+impl Default for IvfFlatIndexBuilder {
+    fn default() -> Self {
+        Self {
+            distance_type: DistanceType::L2,
+            num_partitions: None,
+            sample_rate: 256,
+            max_iterations: 50,
+        }
+    }
+}
+
+impl IvfFlatIndexBuilder {
+    impl_distance_type_setter!();
+    impl_ivf_params_setter!();
+}
+
 /// Builder for an IVF PQ index.
 ///
 /// This index stores a compressed (quantized) copy of every vector.  These vectors

rust/lancedb/src/query.rs

@@ -339,7 +339,7 @@ pub trait QueryBase {
     fn limit(self, limit: usize) -> Self;
 
     /// Set the offset of the query.
-
+    ///
     /// By default, it fetches starting with the first row.
     /// This method can be used to skip the first `offset` rows.
     fn offset(self, offset: usize) -> Self;

rust/lancedb/src/table.rs

@@ -18,9 +18,9 @@ use std::path::Path;
 use std::sync::Arc;
 
 use arrow::array::AsArray;
-use arrow::datatypes::Float32Type;
+use arrow::datatypes::{Float32Type, UInt8Type};
 use arrow_array::{RecordBatchIterator, RecordBatchReader};
-use arrow_schema::{Field, Schema, SchemaRef};
+use arrow_schema::{DataType, Field, Schema, SchemaRef};
 use async_trait::async_trait;
 use datafusion_physical_plan::display::DisplayableExecutionPlan;
 use datafusion_physical_plan::projection::ProjectionExec;
@@ -58,8 +58,8 @@ use crate::embeddings::{EmbeddingDefinition, EmbeddingRegistry, MaybeEmbedded, M
 use crate::error::{Error, Result};
 use crate::index::scalar::FtsIndexBuilder;
 use crate::index::vector::{
-    suggested_num_partitions_for_hnsw, IvfHnswPqIndexBuilder, IvfHnswSqIndexBuilder,
+    suggested_num_partitions_for_hnsw, IvfFlatIndexBuilder, IvfHnswPqIndexBuilder,
-    IvfPqIndexBuilder, VectorIndex,
+    IvfHnswSqIndexBuilder, IvfPqIndexBuilder, VectorIndex,
 };
 use crate::index::IndexStatistics;
 use crate::index::{
@@ -1306,6 +1306,44 @@ impl NativeTable {
             .collect())
     }
 
+    async fn create_ivf_flat_index(
+        &self,
+        index: IvfFlatIndexBuilder,
+        field: &Field,
+        replace: bool,
+    ) -> Result<()> {
+        if !supported_vector_data_type(field.data_type()) {
+            return Err(Error::InvalidInput {
+                message: format!(
+                    "An IVF Flat index cannot be created on the column `{}` which has data type {}",
+                    field.name(),
+                    field.data_type()
+                ),
+            });
+        }
+
+        let num_partitions = if let Some(n) = index.num_partitions {
+            n
+        } else {
+            suggested_num_partitions(self.count_rows(None).await?)
+        };
+        let mut dataset = self.dataset.get_mut().await?;
+        let lance_idx_params = lance::index::vector::VectorIndexParams::ivf_flat(
+            num_partitions as usize,
+            index.distance_type.into(),
+        );
+        dataset
+            .create_index(
+                &[field.name()],
+                IndexType::Vector,
+                None,
+                &lance_idx_params,
+                replace,
+            )
+            .await?;
+        Ok(())
+    }
+
     async fn create_ivf_pq_index(
         &self,
         index: IvfPqIndexBuilder,
@@ -1778,6 +1816,10 @@ impl TableInternal for NativeTable {
             Index::Bitmap(_) => self.create_bitmap_index(field, opts).await,
             Index::LabelList(_) => self.create_label_list_index(field, opts).await,
             Index::FTS(fts_opts) => self.create_fts_index(field, fts_opts, opts.replace).await,
+            Index::IvfFlat(ivf_flat) => {
+                self.create_ivf_flat_index(ivf_flat, field, opts.replace)
+                    .await
+            }
             Index::IvfPq(ivf_pq) => self.create_ivf_pq_index(ivf_pq, field, opts.replace).await,
             Index::IvfHnswPq(ivf_hnsw_pq) => {
                 self.create_ivf_hnsw_pq_index(ivf_hnsw_pq, field, opts.replace)
@@ -1848,8 +1890,14 @@ impl TableInternal for NativeTable {
                 message: format!("Column {} not found in dataset schema", column),
             })?;
 
-            if let arrow_schema::DataType::FixedSizeList(f, dim) = field.data_type() {
+            let mut is_binary = false;
-                if !f.data_type().is_floating() {
+            if let arrow_schema::DataType::FixedSizeList(element, dim) = field.data_type() {
+                match element.data_type() {
+                    e_type if e_type.is_floating() => {}
+                    e_type if *e_type == DataType::UInt8 => {
+                        is_binary = true;
+                    }
+                    _ => {
                         return Err(Error::InvalidInput {
                             message: format!(
                                 "The data type of the vector column '{}' is not a floating point type",
@@ -1857,6 +1905,7 @@ impl TableInternal for NativeTable {
                             ),
                         });
                     }
+                }
                 if dim != query_vector.len() as i32 {
                     return Err(Error::InvalidInput {
                     message: format!(
@@ -1870,6 +1919,15 @@ impl TableInternal for NativeTable {
                 }
             }
 
+            if is_binary {
+                let query_vector = arrow::compute::cast(&query_vector, &DataType::UInt8)?;
+                let query_vector = query_vector.as_primitive::<UInt8Type>();
+                scanner.nearest(
+                    &column,
+                    query_vector,
+                    query.base.limit.unwrap_or(DEFAULT_TOP_K),
+                )?;
+            } else {
                 let query_vector = query_vector.as_primitive::<Float32Type>();
                 scanner.nearest(
                     &column,
@@ -1877,6 +1935,7 @@ impl TableInternal for NativeTable {
                     query.base.limit.unwrap_or(DEFAULT_TOP_K),
                 )?;
             }
+        }
         scanner.limit(
             query.base.limit.map(|limit| limit as i64),
             query.base.offset.map(|offset| offset as i64),

rust/lancedb/src/utils.rs

@@ -110,7 +110,7 @@ pub(crate) fn default_vector_column(schema: &Schema, dim: Option<i32>) -> Result
         .iter()
         .filter_map(|field| match field.data_type() {
             arrow_schema::DataType::FixedSizeList(f, d)
-                if f.data_type().is_floating()
+                if (f.data_type().is_floating() || f.data_type() == &DataType::UInt8)
                     && dim.map(|expect| *d == expect).unwrap_or(true) =>
             {
                 Some(field.name())
@@ -171,7 +171,9 @@ pub fn supported_fts_data_type(dtype: &DataType) -> bool {
 
 pub fn supported_vector_data_type(dtype: &DataType) -> bool {
     match dtype {
-        DataType::FixedSizeList(inner, _) => DataType::is_floating(inner.data_type()),
+        DataType::FixedSizeList(inner, _) => {
+            DataType::is_floating(inner.data_type()) || *inner.data_type() == DataType::UInt8
+        }
         _ => false,
     }
 }