test

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

.github/workflows/docs.yml

@@ -58,7 +58,7 @@ jobs:
           cache: 'npm'
           cache-dependency-path: docs/package-lock.json
       - name: Install node dependencies
-        working-directory: nodejs
+        working-directory: node
         run: |
           sudo apt update
           sudo apt install -y protobuf-compiler libssl-dev

Cargo.toml

@@ -23,7 +23,7 @@ lance-table = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https:
 lance-testing = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }
 lance-datafusion = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }
 lance-encoding = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }
-lance-namespace = "0.0.16"
+lance-namespace = "0.0.15"
 # Note that this one does not include pyarrow
 arrow = { version = "55.1", optional = false }
 arrow-array = "55.1"
@@ -31,6 +31,7 @@ arrow-data = "55.1"
 arrow-ipc = "55.1"
 arrow-ord = "55.1"
 arrow-schema = "55.1"
+arrow-arith = "55.1"
 arrow-cast = "55.1"
 async-trait = "0"
 datafusion = { version = "49.0", default-features = false }
@@ -51,6 +52,7 @@ pin-project = "1.0.7"
 snafu = "0.8"
 url = "2"
 num-traits = "0.2"
+rand = "0.9"
 regex = "1.10"
 lazy_static = "1"
 semver = "1.0.25"
@@ -58,16 +60,7 @@ crunchy = "0.2.4"
 # Temporary pins to work around downstream issues
 # https://github.com/apache/arrow-rs/commit/2fddf85afcd20110ce783ed5b4cdeb82293da30b
 chrono = "=0.4.41"
+# https://github.com/RustCrypto/formats/issues/1684
+base64ct = "=1.6.0"
 # Workaround for: https://github.com/Lokathor/bytemuck/issues/306
 bytemuck_derive = ">=1.8.1, <1.9.0"
-
-[patch.crates-io]
-# Force to use the same lance version as the rest of the project to avoid duplicate dependencies
-lance = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }
-lance-io = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }
-lance-index = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }
-lance-linalg = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }
-lance-table = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }
-lance-testing = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }
-lance-datafusion = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }
-lance-encoding = { "version" = "=0.37.0", "tag" = "v0.37.1-beta.1", "git" = "https://github.com/lancedb/lance.git" }

ci/run_with_test_connection.sh

@@ -16,47 +16,30 @@ check_command_exists() {
 }
 
 if [[ ! -e ./lancedb ]]; then
-    if [[ -v SOPHON_READ_TOKEN ]]; then
-        INPUT="lancedb-linux-x64"
-        gh release \
-            --repo lancedb/lancedb \
-            download ci-support-binaries \
-            --pattern "${INPUT}" \
-            || die "failed to fetch cli."
-        check_command_exists openssl
-        openssl enc -aes-256-cbc \
-            -d -pbkdf2 \
-            -pass "env:SOPHON_READ_TOKEN" \
-            -in "${INPUT}" \
-            -out ./lancedb-linux-x64.tar.gz \
-            || die "openssl failed"
-        TARGET="${INPUT}.tar.gz"
-    else
-        ARCH="x64"
-        if [[ $OSTYPE == 'darwin'* ]]; then
-            UNAME=$(uname -m)
-            if [[ $UNAME == 'arm64' ]]; then
-                ARCH='arm64'
-            fi
-            OSTYPE="macos"
-        elif [[ $OSTYPE == 'linux'* ]]; then
-            if [[ $UNAME == 'aarch64' ]]; then
-                ARCH='arm64'
-            fi
-            OSTYPE="linux"
-        else
-            die "unknown OSTYPE: $OSTYPE"
+    ARCH="x64"
+    if [[ $OSTYPE == 'darwin'* ]]; then
+        UNAME=$(uname -m)
+        if [[ $UNAME == 'arm64' ]]; then
+            ARCH='arm64'
         fi
-
-        check_command_exists gh
-        TARGET="lancedb-${OSTYPE}-${ARCH}.tar.gz"
-        gh release \
-            --repo lancedb/sophon \
-            download lancedb-cli-v0.0.3 \
-            --pattern "${TARGET}" \
-            || die "failed to fetch cli."
+        OSTYPE="macos"
+    elif [[ $OSTYPE == 'linux'* ]]; then
+        if [[ $UNAME == 'aarch64' ]]; then
+            ARCH='arm64'
+        fi
+        OSTYPE="linux"
+    else
+        die "unknown OSTYPE: $OSTYPE"
     fi
 
+    check_command_exists gh
+    TARGET="lancedb-${OSTYPE}-${ARCH}.tar.gz"
+    gh release \
+        --repo lancedb/sophon \
+        download lancedb-cli-v0.0.3 \
+        --pattern "${TARGET}" \
+        || die "failed to fetch cli."
+
     check_command_exists tar
     tar xvf "${TARGET}" || die "tar failed."
     [[ -e ./lancedb ]] || die "failed to extract lancedb."

ci/set_lance_version.py

@@ -117,7 +117,7 @@ def update_cargo_toml(line_updater):
     lance_line = ""
     is_parsing_lance_line = False
     for line in lines:
-        if line.startswith("lance") and not line.startswith("lance-namespace"):
+        if line.startswith("lance"):
             # Check if this is a single-line or multi-line entry
             # Single-line entries either:
             # 1. End with } (complete inline table)

docs/src/js/classes/Index.md

@@ -194,6 +194,37 @@ currently is also a memory intensive operation.
 
 ***
 
+### ivfRq()
+
+```ts
+static ivfRq(options?): Index
+```
+
+Create an IvfRq index
+
+IVF-RQ (RabitQ Quantization) compresses vectors using RabitQ quantization
+and organizes them into IVF partitions.
+
+The compression scheme is called RabitQ quantization. Each dimension is quantized into a small number of bits.
+The parameters `num_bits` and `num_partitions` control this process, providing a tradeoff
+between index size (and thus search speed) and index accuracy.
+
+The partitioning process is called IVF and the `num_partitions` parameter controls how
+many groups to create.
+
+Note that training an IVF RQ index on a large dataset is a slow operation and
+currently is also a memory intensive operation.
+
+#### Parameters
+
+* **options?**: `Partial`<[`IvfRqOptions`](../interfaces/IvfRqOptions.md)>
+
+#### Returns
+
+[`Index`](Index.md)
+
+***
+
 ### labelList()
 
 ```ts

docs/src/js/classes/MergeInsertBuilder.md

@@ -52,6 +52,30 @@ the merge result
 
 ***
 
+### useIndex()
+
+```ts
+useIndex(useIndex): MergeInsertBuilder
+```
+
+Controls whether to use indexes for the merge operation.
+
+When set to `true` (the default), the operation will use an index if available
+on the join key for improved performance. When set to `false`, it forces a full
+table scan even if an index exists. This can be useful for benchmarking or when
+the query optimizer chooses a suboptimal path.
+
+#### Parameters
+
+* **useIndex**: `boolean`
+    Whether to use indices for the merge operation. Defaults to `true`.
+
+#### Returns
+
+[`MergeInsertBuilder`](MergeInsertBuilder.md)
+
+***
+
 ### whenMatchedUpdateAll()
 
 ```ts

docs/src/js/globals.md

@@ -68,6 +68,7 @@
 - [IndexStatistics](interfaces/IndexStatistics.md)
 - [IvfFlatOptions](interfaces/IvfFlatOptions.md)
 - [IvfPqOptions](interfaces/IvfPqOptions.md)
+- [IvfRqOptions](interfaces/IvfRqOptions.md)
 - [MergeResult](interfaces/MergeResult.md)
 - [OpenTableOptions](interfaces/OpenTableOptions.md)
 - [OptimizeOptions](interfaces/OptimizeOptions.md)

nodejs/__test__/table.test.ts

@@ -804,6 +804,15 @@ describe("When creating an index", () => {
     });
   });
 
+  it("should be able to create IVF_RQ", async () => {
+    await tbl.createIndex("vec", {
+      config: Index.ivfRq({
+        numPartitions: 10,
+        numBits: 1,
+      }),
+    });
+  });
+
   it("should allow me to replace (or not) an existing index", async () => {
     await tbl.createIndex("id");
     // Default is replace=true

nodejs/lancedb/index.ts

@@ -85,6 +85,7 @@ export {
   Index,
   IndexOptions,
   IvfPqOptions,
+  IvfRqOptions,
   IvfFlatOptions,
   HnswPqOptions,
   HnswSqOptions,

nodejs/lancedb/indices.ts

@@ -112,6 +112,77 @@ export interface IvfPqOptions {
   sampleRate?: number;
 }
 
+export interface IvfRqOptions {
+  /**
+   * 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.
+   */
+  numPartitions?: number;
+
+  /**
+   * Number of bits per dimension for residual quantization.
+   *
+   * This value controls how much each residual component is compressed. The more
+   * bits, the more accurate the index will be but the slower search. Typical values
+   * are small integers; the default is 1 bit per dimension.
+   */
+  numBits?: number;
+
+  /**
+   * Distance type to use to build the index.
+   *
+   * Default value is "l2".
+   *
+   * 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 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.
+   * "cosine" - Cosine distance.
+   * "dot" - Dot product.
+   */
+  distanceType?: "l2" | "cosine" | "dot";
+
+  /**
+   * Max iterations to train IVF kmeans.
+   *
+   * When training an IVF index we use kmeans to calculate the partitions. This parameter
+   * controls how many iterations of kmeans to run.
+   *
+   * The default value is 50.
+   */
+  maxIterations?: number;
+
+  /**
+   * The number of vectors, per partition, to sample when training IVF kmeans.
+   *
+   * When an IVF 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.
+   */
+  sampleRate?: number;
+}
+
 /**
  * Options to create an `HNSW_PQ` index
  */
@@ -523,6 +594,35 @@ export class Index {
         options?.distanceType,
         options?.numPartitions,
         options?.numSubVectors,
+        options?.numBits,
+        options?.maxIterations,
+        options?.sampleRate,
+      ),
+    );
+  }
+
+  /**
+   * Create an IvfRq index
+   *
+   * IVF-RQ (RabitQ Quantization) compresses vectors using RabitQ quantization
+   * and organizes them into IVF partitions.
+   *
+   * The compression scheme is called RabitQ quantization. Each dimension is quantized into a small number of bits.
+   * The parameters `num_bits` and `num_partitions` control this process, providing a tradeoff
+   * between index size (and thus search speed) and index accuracy.
+   *
+   * The partitioning process is called IVF and the `num_partitions` parameter controls how
+   * many groups to create.
+   *
+   * Note that training an IVF RQ index on a large dataset is a slow operation and
+   * currently is also a memory intensive operation.
+   */
+  static ivfRq(options?: Partial<IvfRqOptions>) {
+    return new Index(
+      LanceDbIndex.ivfRq(
+        options?.distanceType,
+        options?.numPartitions,
+        options?.numBits,
         options?.maxIterations,
         options?.sampleRate,
       ),

nodejs/src/index.rs

@@ -6,6 +6,7 @@ use std::sync::Mutex;
 use lancedb::index::scalar::{BTreeIndexBuilder, FtsIndexBuilder};
 use lancedb::index::vector::{
     IvfFlatIndexBuilder, IvfHnswPqIndexBuilder, IvfHnswSqIndexBuilder, IvfPqIndexBuilder,
+    IvfRqIndexBuilder,
 };
 use lancedb::index::Index as LanceDbIndex;
 use napi_derive::napi;
@@ -65,6 +66,36 @@ impl Index {
         })
     }
 
+    #[napi(factory)]
+    pub fn ivf_rq(
+        distance_type: Option<String>,
+        num_partitions: Option<u32>,
+        num_bits: Option<u32>,
+        max_iterations: Option<u32>,
+        sample_rate: Option<u32>,
+    ) -> napi::Result<Self> {
+        let mut ivf_rq_builder = IvfRqIndexBuilder::default();
+        if let Some(distance_type) = distance_type {
+            let distance_type = parse_distance_type(distance_type)?;
+            ivf_rq_builder = ivf_rq_builder.distance_type(distance_type);
+        }
+        if let Some(num_partitions) = num_partitions {
+            ivf_rq_builder = ivf_rq_builder.num_partitions(num_partitions);
+        }
+        if let Some(num_bits) = num_bits {
+            ivf_rq_builder = ivf_rq_builder.num_bits(num_bits);
+        }
+        if let Some(max_iterations) = max_iterations {
+            ivf_rq_builder = ivf_rq_builder.max_iterations(max_iterations);
+        }
+        if let Some(sample_rate) = sample_rate {
+            ivf_rq_builder = ivf_rq_builder.sample_rate(sample_rate);
+        }
+        Ok(Self {
+            inner: Mutex::new(Some(LanceDbIndex::IvfRq(ivf_rq_builder))),
+        })
+    }
+
     #[napi(factory)]
     pub fn ivf_flat(
         distance_type: Option<String>,

python/.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.25.2-beta.1"
+current_version = "0.25.2-beta.0"
 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 = "0.25.2-beta.1"
+version = "0.25.2-beta.0"
 edition.workspace = true
 description = "Python bindings for LanceDB"
 license.workspace = true

python/pyproject.toml

@@ -10,7 +10,7 @@ dependencies = [
     "pyarrow>=16",
     "pydantic>=1.10",
     "tqdm>=4.27.0",
-    "lance-namespace>=0.0.16"
+    "lance-namespace==0.0.6"
 ]
 description = "lancedb"
 authors = [{ name = "LanceDB Devs", email = "dev@lancedb.com" }]

python/python/lancedb/index.py

@@ -605,9 +605,53 @@ class IvfPq:
     target_partition_size: Optional[int] = None
 
 
+@dataclass
+class IvfRq:
+    """Describes an IVF RQ Index
+
+    IVF-RQ (Residual Quantization) stores a compressed copy of each vector using
+    residual quantization and organizes them into IVF partitions. Parameters
+    largely mirror IVF-PQ for consistency.
+
+    Attributes
+    ----------
+    distance_type: str, default "l2"
+        Distance metric used to train the index and for quantization.
+
+        The following distance types are available:
+
+        "l2" - Euclidean distance.
+        "cosine" - Cosine distance.
+        "dot" - Dot product.
+
+    num_partitions: int, default sqrt(num_rows)
+        Number of IVF partitions to create.
+
+    num_bits: int, default 1
+        Number of bits to encode each dimension.
+
+    max_iterations: int, default 50
+        Max iterations to train kmeans when computing IVF partitions.
+
+    sample_rate: int, default 256
+        Controls the number of training vectors: sample_rate * num_partitions.
+
+    target_partition_size, default is 8192
+        Target size of each partition.
+    """
+
+    distance_type: Literal["l2", "cosine", "dot"] = "l2"
+    num_partitions: Optional[int] = None
+    num_bits: int = 1
+    max_iterations: int = 50
+    sample_rate: int = 256
+    target_partition_size: Optional[int] = None
+
+
 __all__ = [
     "BTree",
     "IvfPq",
+    "IvfRq",
     "IvfFlat",
     "HnswPq",
     "HnswSq",

python/python/lancedb/table.py

@@ -44,7 +44,7 @@ import numpy as np
 
 from .common import DATA, VEC, VECTOR_COLUMN_NAME
 from .embeddings import EmbeddingFunctionConfig, EmbeddingFunctionRegistry
-from .index import BTree, IvfFlat, IvfPq, Bitmap, LabelList, HnswPq, HnswSq, FTS
+from .index import BTree, IvfFlat, IvfPq, Bitmap, IvfRq, LabelList, HnswPq, HnswSq, FTS
 from .merge import LanceMergeInsertBuilder
 from .pydantic import LanceModel, model_to_dict
 from .query import (
@@ -1991,7 +1991,7 @@ class LanceTable(Table):
         index_cache_size: Optional[int] = None,
         num_bits: int = 8,
         index_type: Literal[
-            "IVF_FLAT", "IVF_PQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ"
+            "IVF_FLAT", "IVF_PQ", "IVF_RQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ"
         ] = "IVF_PQ",
         max_iterations: int = 50,
         sample_rate: int = 256,
@@ -2039,6 +2039,15 @@ class LanceTable(Table):
                 sample_rate=sample_rate,
                 target_partition_size=target_partition_size,
             )
+        elif index_type == "IVF_RQ":
+            config = IvfRq(
+                distance_type=metric,
+                num_partitions=num_partitions,
+                num_bits=num_bits,
+                max_iterations=max_iterations,
+                sample_rate=sample_rate,
+                target_partition_size=target_partition_size,
+            )
         elif index_type == "IVF_HNSW_PQ":
             config = HnswPq(
                 distance_type=metric,
@@ -3330,7 +3339,7 @@ class AsyncTable:
         *,
         replace: Optional[bool] = None,
         config: Optional[
-            Union[IvfFlat, IvfPq, HnswPq, HnswSq, BTree, Bitmap, LabelList, FTS]
+            Union[IvfFlat, IvfPq, IvfRq, HnswPq, HnswSq, BTree, Bitmap, LabelList, FTS]
         ] = None,
         wait_timeout: Optional[timedelta] = None,
         name: Optional[str] = None,
@@ -3369,11 +3378,12 @@ class AsyncTable:
         """
         if config is not None:
             if not isinstance(
-                config, (IvfFlat, IvfPq, HnswPq, HnswSq, BTree, Bitmap, LabelList, FTS)
+                config,
+                (IvfFlat, IvfPq, IvfRq, HnswPq, HnswSq, BTree, Bitmap, LabelList, FTS),
             ):
                 raise TypeError(
-                    "config must be an instance of IvfPq, HnswPq, HnswSq, BTree,"
-                    " Bitmap, LabelList, or FTS"
+                    "config must be an instance of IvfPq, IvfRq, HnswPq, HnswSq, BTree,"
+                    " Bitmap, LabelList, or FTS, but got " + str(type(config))
                 )
         try:
             await self._inner.create_index(

python/python/lancedb/types.py

@@ -18,10 +18,17 @@ AddMode = Literal["append", "overwrite"]
 CreateMode = Literal["create", "overwrite"]
 
 # Index type literals
-VectorIndexType = Literal["IVF_FLAT", "IVF_PQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ"]
+VectorIndexType = Literal["IVF_FLAT", "IVF_PQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ", "IVF_RQ"]
 ScalarIndexType = Literal["BTREE", "BITMAP", "LABEL_LIST"]
 IndexType = Literal[
-    "IVF_PQ", "IVF_HNSW_PQ", "IVF_HNSW_SQ", "FTS", "BTREE", "BITMAP", "LABEL_LIST"
+    "IVF_PQ",
+    "IVF_HNSW_PQ",
+    "IVF_HNSW_SQ",
+    "FTS",
+    "BTREE",
+    "BITMAP",
+    "LABEL_LIST",
+    "IVF_RQ",
 ]
 
 # Tokenizer literals

python/python/tests/test_index.py

@@ -8,7 +8,17 @@ import pyarrow as pa
 import pytest
 import pytest_asyncio
 from lancedb import AsyncConnection, AsyncTable, connect_async
-from lancedb.index import BTree, IvfFlat, IvfPq, Bitmap, LabelList, HnswPq, HnswSq, FTS
+from lancedb.index import (
+    BTree,
+    IvfFlat,
+    IvfPq,
+    IvfRq,
+    Bitmap,
+    LabelList,
+    HnswPq,
+    HnswSq,
+    FTS,
+)
 
 
 @pytest_asyncio.fixture
@@ -195,6 +205,16 @@ async def test_create_4bit_ivfpq_index(some_table: AsyncTable):
     assert stats.loss >= 0.0
 
 
+@pytest.mark.asyncio
+async def test_create_ivfrq_index(some_table: AsyncTable):
+    await some_table.create_index("vector", config=IvfRq(num_bits=1))
+    indices = await some_table.list_indices()
+    assert len(indices) == 1
+    assert indices[0].index_type == "IvfRq"
+    assert indices[0].columns == ["vector"]
+    assert indices[0].name == "vector_idx"
+
+
 @pytest.mark.asyncio
 async def test_create_hnswpq_index(some_table: AsyncTable):
     await some_table.create_index("vector", config=HnswPq(num_partitions=10))

python/src/index.rs

@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
-use lancedb::index::vector::IvfFlatIndexBuilder;
+use lancedb::index::vector::{IvfFlatIndexBuilder, IvfRqIndexBuilder};
 use lancedb::index::{
     scalar::{BTreeIndexBuilder, FtsIndexBuilder},
     vector::{IvfHnswPqIndexBuilder, IvfHnswSqIndexBuilder, IvfPqIndexBuilder},
@@ -87,6 +87,22 @@ pub fn extract_index_params(source: &Option<Bound<'_, PyAny>>) -> PyResult<Lance
                 }
                 Ok(LanceDbIndex::IvfPq(ivf_pq_builder))
             },
+            "IvfRq" => {
+                let params = source.extract::<IvfRqParams>()?;
+                let distance_type = parse_distance_type(params.distance_type)?;
+                let mut ivf_rq_builder = IvfRqIndexBuilder::default()
+                    .distance_type(distance_type)
+                    .max_iterations(params.max_iterations)
+                    .sample_rate(params.sample_rate)
+                    .num_bits(params.num_bits);
+                if let Some(num_partitions) = params.num_partitions {
+                    ivf_rq_builder = ivf_rq_builder.num_partitions(num_partitions);
+                }
+                if let Some(target_partition_size) = params.target_partition_size {
+                    ivf_rq_builder = ivf_rq_builder.target_partition_size(target_partition_size);
+                }
+                Ok(LanceDbIndex::IvfRq(ivf_rq_builder))
+            },
             "HnswPq" => {
                 let params = source.extract::<IvfHnswPqParams>()?;
                 let distance_type = parse_distance_type(params.distance_type)?;
@@ -170,6 +186,16 @@ struct IvfPqParams {
     target_partition_size: Option<u32>,
 }
 
+#[derive(FromPyObject)]
+struct IvfRqParams {
+    distance_type: String,
+    num_partitions: Option<u32>,
+    num_bits: u32,
+    max_iterations: u32,
+    sample_rate: u32,
+    target_partition_size: Option<u32>,
+}
+
 #[derive(FromPyObject)]
 struct IvfHnswPqParams {
     distance_type: String,

rust/lancedb/src/index.rs

@@ -8,6 +8,7 @@ use std::sync::Arc;
 use std::time::Duration;
 use vector::IvfFlatIndexBuilder;
 
+use crate::index::vector::IvfRqIndexBuilder;
 use crate::{table::BaseTable, DistanceType, Error, Result};
 
 use self::{
@@ -53,6 +54,9 @@ pub enum Index {
     /// IVF index with Product Quantization
     IvfPq(IvfPqIndexBuilder),
 
+    /// IVF index with RabitQ Quantization
+    IvfRq(IvfRqIndexBuilder),
+
     /// IVF-HNSW index with Product Quantization
     /// It is a variant of the HNSW algorithm that uses product quantization to compress the vectors.
     IvfHnswPq(IvfHnswPqIndexBuilder),
@@ -275,6 +279,8 @@ pub enum IndexType {
     IvfFlat,
     #[serde(alias = "IVF_PQ")]
     IvfPq,
+    #[serde(alias = "IVF_RQ")]
+    IvfRq,
     #[serde(alias = "IVF_HNSW_PQ")]
     IvfHnswPq,
     #[serde(alias = "IVF_HNSW_SQ")]
@@ -296,6 +302,7 @@ impl std::fmt::Display for IndexType {
         match self {
             Self::IvfFlat => write!(f, "IVF_FLAT"),
             Self::IvfPq => write!(f, "IVF_PQ"),
+            Self::IvfRq => write!(f, "IVF_RQ"),
             Self::IvfHnswPq => write!(f, "IVF_HNSW_PQ"),
             Self::IvfHnswSq => write!(f, "IVF_HNSW_SQ"),
             Self::BTree => write!(f, "BTREE"),
@@ -317,6 +324,7 @@ impl std::str::FromStr for IndexType {
             "FTS" | "INVERTED" => Ok(Self::FTS),
             "IVF_FLAT" => Ok(Self::IvfFlat),
             "IVF_PQ" => Ok(Self::IvfPq),
+            "IVF_RQ" => Ok(Self::IvfRq),
             "IVF_HNSW_PQ" => Ok(Self::IvfHnswPq),
             "IVF_HNSW_SQ" => Ok(Self::IvfHnswSq),
             _ => Err(Error::InvalidInput {

rust/lancedb/src/index/vector.rs

@@ -291,6 +291,52 @@ pub(crate) fn suggested_num_sub_vectors(dim: u32) -> u32 {
     }
 }
 
+/// Builder for an IVF RQ index.
+///
+/// This index stores a compressed (quantized) copy of every vector. Each dimension
+/// is quantized into a small number of bits.
+/// The parameters `num_bits` control this process, providing a tradeoff
+/// between index size (and thus search speed) and index accuracy.
+///
+/// The partitioning process is called IVF and the `num_partitions` parameter controls how
+/// many groups to create.
+///
+/// Note that training an IVF RQ index on a large dataset is a slow operation and
+/// currently is also a memory intensive operation.
+#[derive(Debug, Clone)]
+pub struct IvfRqIndexBuilder {
+    // IVF
+    pub(crate) distance_type: DistanceType,
+    pub(crate) num_partitions: Option<u32>,
+    pub(crate) num_bits: Option<u32>,
+    pub(crate) sample_rate: u32,
+    pub(crate) max_iterations: u32,
+    pub(crate) target_partition_size: Option<u32>,
+}
+
+impl Default for IvfRqIndexBuilder {
+    fn default() -> Self {
+        Self {
+            distance_type: DistanceType::L2,
+            num_partitions: None,
+            num_bits: None,
+            sample_rate: 256,
+            max_iterations: 50,
+            target_partition_size: None,
+        }
+    }
+}
+
+impl IvfRqIndexBuilder {
+    impl_distance_type_setter!();
+    impl_ivf_params_setter!();
+
+    pub fn num_bits(mut self, num_bits: u32) -> Self {
+        self.num_bits = Some(num_bits);
+        self
+    }
+}
+
 /// Builder for an IVF HNSW PQ index.
 ///
 /// This index is a combination of IVF and HNSW.

rust/lancedb/src/table.rs

@@ -1838,6 +1838,18 @@ impl NativeTable {
                 );
                 Ok(Box::new(lance_idx_params))
             }
+            Index::IvfRq(index) => {
+                Self::validate_index_type(field, "IVF RQ", supported_vector_data_type)?;
+                let num_partitions = self
+                    .get_num_partitions(index.num_partitions, false, None)
+                    .await?;
+                let lance_idx_params = VectorIndexParams::ivf_rq(
+                    num_partitions as usize,
+                    index.num_bits.unwrap_or(1) as u8,
+                    index.distance_type.into(),
+                );
+                Ok(Box::new(lance_idx_params))
+            }
             Index::IvfHnswPq(index) => {
                 Self::validate_index_type(field, "IVF HNSW PQ", supported_vector_data_type)?;
                 let dim = Self::get_vector_dimension(field)?;
@@ -1907,9 +1919,11 @@ impl NativeTable {
             Index::Bitmap(_) => IndexType::Bitmap,
             Index::LabelList(_) => IndexType::LabelList,
             Index::FTS(_) => IndexType::Inverted,
-            Index::IvfFlat(_) | Index::IvfPq(_) | Index::IvfHnswPq(_) | Index::IvfHnswSq(_) => {
-                IndexType::Vector
-            }
+            Index::IvfFlat(_)
+            | Index::IvfPq(_)
+            | Index::IvfRq(_)
+            | Index::IvfHnswPq(_)
+            | Index::IvfHnswSq(_) => IndexType::Vector,
         }
     }
 

rust/lancedb/src/utils.rs

@@ -39,7 +39,7 @@ impl PatchStoreParam for Option<ObjectStoreParams> {
         let mut params = self.unwrap_or_default();
         if params.object_store_wrapper.is_some() {
             return Err(Error::Other {
-                message: "can not patch param because object store is already set".into(),
+                message: "can not patch param because object store is already set.".into(),
                 source: None,
             });
         }