Bump version: 0.9.0-beta.0 → 0.9.0

PR fixes #1514

.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.7.2"
+current_version = "0.9.0"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

.github/workflows/java.yml

@@ -3,6 +3,8 @@ on:
   push:
     branches:
       - main
+    paths:
+      - java/**
   pull_request:
     paths:
       - java/**
@@ -21,9 +23,42 @@ env:
   CARGO_INCREMENTAL: "0"
   CARGO_BUILD_JOBS: "1"
 jobs:
-  linux-build:
+  linux-build-java-11:
     runs-on: ubuntu-22.04
-    name: ubuntu-22.04 + Java 11 & 17
+    name: ubuntu-22.04 + Java 11
+    defaults:
+      run:
+        working-directory: ./java
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+      - uses: Swatinem/rust-cache@v2
+        with:
+          workspaces: java/core/lancedb-jni
+      - name: Run cargo fmt
+        run: cargo fmt --check
+        working-directory: ./java/core/lancedb-jni
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y protobuf-compiler libssl-dev
+      - name: Install Java 11
+        uses: actions/setup-java@v4
+        with:
+          distribution: temurin
+          java-version: 11
+          cache: "maven"
+      - name: Java Style Check
+        run: mvn checkstyle:check
+      # Disable because of issues in lancedb rust core code
+      # - name: Rust Clippy
+      #   working-directory: java/core/lancedb-jni
+      #   run: cargo clippy --all-targets -- -D warnings
+      - name: Running tests with Java 11
+        run: mvn clean test
+  linux-build-java-17:
+    runs-on: ubuntu-22.04
+    name: ubuntu-22.04 + Java 17
     defaults:
       run:
         working-directory: ./java
@@ -47,20 +82,12 @@ jobs:
           java-version: 17
           cache: "maven"
       - run: echo "JAVA_17=$JAVA_HOME" >> $GITHUB_ENV
-      - name: Install Java 11
-        uses: actions/setup-java@v4
-        with:
-          distribution: temurin
-          java-version: 11
-          cache: "maven"
       - name: Java Style Check
         run: mvn checkstyle:check
       # Disable because of issues in lancedb rust core code
       # - name: Rust Clippy
       #   working-directory: java/core/lancedb-jni
       #   run: cargo clippy --all-targets -- -D warnings
-      - name: Running tests with Java 11
-        run: mvn clean test
       - name: Running tests with Java 17
         run: |
           export JAVA_TOOL_OPTIONS="$JAVA_TOOL_OPTIONS \
@@ -83,3 +110,4 @@ jobs:
           -Djdk.reflect.useDirectMethodHandle=false \
           -Dio.netty.tryReflectionSetAccessible=true"
           JAVA_HOME=$JAVA_17 mvn clean test
+  

Cargo.toml

@@ -20,20 +20,21 @@ keywords = ["lancedb", "lance", "database", "vector", "search"]
 categories = ["database-implementations"]
 
 [workspace.dependencies]
-lance = { "version" = "=0.15.0", "features" = ["dynamodb"] }
-lance-index = { "version" = "=0.15.0" }
-lance-linalg = { "version" = "=0.15.0" }
-lance-testing = { "version" = "=0.15.0" }
-lance-datafusion = { "version" = "=0.15.0" }
+lance = { "version" = "=0.16.0", "features" = ["dynamodb"] }
+lance-index = { "version" = "=0.16.0" }
+lance-linalg = { "version" = "=0.16.0" }
+lance-testing = { "version" = "=0.16.0" }
+lance-datafusion = { "version" = "=0.16.0" }
+lance-encoding = { "version" = "=0.16.0" }
 # Note that this one does not include pyarrow
-arrow = { version = "52.1", optional = false }
-arrow-array = "52.1"
-arrow-data = "52.1"
-arrow-ipc = "52.1"
-arrow-ord = "52.1"
-arrow-schema = "52.1"
-arrow-arith = "52.1"
-arrow-cast = "52.1"
+arrow = { version = "52.2", optional = false }
+arrow-array = "52.2"
+arrow-data = "52.2"
+arrow-ipc = "52.2"
+arrow-ord = "52.2"
+arrow-schema = "52.2"
+arrow-arith = "52.2"
+arrow-cast = "52.2"
 async-trait = "0"
 chrono = "0.4.35"
 datafusion-physical-plan = "40.0"

ci/build_linux_artifacts.sh

@@ -18,4 +18,4 @@ docker run \
     -v $(pwd):/io -w /io \
     --memory-swap=-1 \
     lancedb-node-manylinux \
-    bash ci/manylinux_node/build.sh $ARCH
+    bash ci/manylinux_node/build_vectordb.sh $ARCH

ci/build_linux_artifacts_nodejs.sh

@@ -4,9 +4,9 @@ ARCH=${1:-x86_64}
 
 # We pass down the current user so that when we later mount the local files
 # into the container, the files are accessible by the current user.
-pushd ci/manylinux_nodejs
+pushd ci/manylinux_node
 docker build \
-    -t lancedb-nodejs-manylinux \
+    -t lancedb-node-manylinux-$ARCH \
     --build-arg="ARCH=$ARCH" \
     --build-arg="DOCKER_USER=$(id -u)" \
     --progress=plain \
@@ -17,5 +17,5 @@ popd
 docker run \
     -v $(pwd):/io -w /io \
     --memory-swap=-1 \
-    lancedb-nodejs-manylinux \
-    bash ci/manylinux_nodejs/build.sh $ARCH
+    lancedb-node-manylinux-$ARCH \
+    bash ci/manylinux_node/build_lancedb.sh $ARCH

ci/manylinux_node/Dockerfile

@@ -4,7 +4,7 @@
 # range of linux distributions.
 ARG ARCH=x86_64
 
-FROM quay.io/pypa/manylinux2014_${ARCH}
+FROM quay.io/pypa/manylinux_2_28_${ARCH}
 
 ARG ARCH=x86_64
 ARG DOCKER_USER=default_user

ci/manylinux_node/install_openssl.sh

@@ -6,7 +6,7 @@
 # /usr/bin/ld: failed to set dynamic section sizes: Bad value
 set -e
 
-git clone -b OpenSSL_1_1_1u \
+git clone -b OpenSSL_1_1_1v \
     --single-branch \
     https://github.com/openssl/openssl.git
 

ci/manylinux_node/prepare_manylinux_node.sh

@@ -8,7 +8,7 @@ install_node() {
 
     source "$HOME"/.bashrc
 
-    nvm install --no-progress 16
+    nvm install --no-progress 18
 }
 
 install_rust() {

ci/manylinux_nodejs/Dockerfile

@@ -1,31 +0,0 @@
-# Many linux dockerfile with Rust, Node, and Lance dependencies installed.
-# This container allows building the node modules native libraries in an
-# environment with a very old glibc, so that we are compatible with a wide
-# range of linux distributions.
-ARG ARCH=x86_64
-
-FROM quay.io/pypa/manylinux2014_${ARCH}
-
-ARG ARCH=x86_64
-ARG DOCKER_USER=default_user
-
-# Install static openssl
-COPY install_openssl.sh install_openssl.sh
-RUN ./install_openssl.sh ${ARCH} > /dev/null
-
-# Protobuf is also installed as root.
-COPY install_protobuf.sh install_protobuf.sh
-RUN ./install_protobuf.sh ${ARCH}
-
-ENV DOCKER_USER=${DOCKER_USER}
-# Create a group and user
-RUN echo ${ARCH} && adduser --user-group --create-home --uid ${DOCKER_USER} build_user
-
-# We switch to the user to install Rust and Node, since those like to be
-# installed at the user level.
-USER ${DOCKER_USER}
-
-COPY prepare_manylinux_node.sh prepare_manylinux_node.sh
-RUN cp /prepare_manylinux_node.sh $HOME/ && \
-    cd $HOME && \
-    ./prepare_manylinux_node.sh ${ARCH}

ci/manylinux_nodejs/install_openssl.sh

@@ -1,26 +0,0 @@
-#!/bin/bash
-# Builds openssl from source so we can statically link to it
-
-# this is to avoid the error we get with the system installation:
-# /usr/bin/ld: <library>: version node not found for symbol SSLeay@@OPENSSL_1.0.1
-# /usr/bin/ld: failed to set dynamic section sizes: Bad value
-set -e
-
-git clone -b OpenSSL_1_1_1u \
-    --single-branch \
-    https://github.com/openssl/openssl.git
-
-pushd openssl
-
-if [[ $1 == x86_64* ]]; then
-    ARCH=linux-x86_64
-else
-    # gnu target
-    ARCH=linux-aarch64
-fi
-
-./Configure no-shared $ARCH
-
-make
-
-make install

ci/manylinux_nodejs/install_protobuf.sh

@@ -1,15 +0,0 @@
-#!/bin/bash
-# Installs protobuf compiler. Should be run as root.
-set -e
-
-if [[ $1 == x86_64* ]]; then
-    ARCH=x86_64
-else
-    # gnu target
-    ARCH=aarch_64
-fi
-
-PB_REL=https://github.com/protocolbuffers/protobuf/releases
-PB_VERSION=23.1
-curl -LO $PB_REL/download/v$PB_VERSION/protoc-$PB_VERSION-linux-$ARCH.zip
-unzip protoc-$PB_VERSION-linux-$ARCH.zip -d /usr/local

ci/manylinux_nodejs/prepare_manylinux_node.sh

@@ -1,21 +0,0 @@
-#!/bin/bash
-set -e
-
-install_node() {
-    echo "Installing node..."
-
-    curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.34.0/install.sh | bash
-
-    source "$HOME"/.bashrc
-
-    nvm install --no-progress 16
-}
-
-install_rust() {
-    echo "Installing rust..."
-    curl https://sh.rustup.rs -sSf | bash -s -- -y
-    export PATH="$PATH:/root/.cargo/bin"
-}
-
-install_node
-install_rust

docs/mkdocs.yml

@@ -141,12 +141,15 @@ nav:
           - Overview: examples/index.md
           - 🐍 Python:
               - Overview: examples/examples_python.md
-              - YouTube Transcript Search: notebooks/youtube_transcript_search.ipynb
-              - Documentation QA Bot using LangChain: notebooks/code_qa_bot.ipynb
-              - Multimodal search using CLIP: notebooks/multimodal_search.ipynb
-              - Example - Calculate CLIP Embeddings with Roboflow Inference: examples/image_embeddings_roboflow.md
-              - Serverless QA Bot with S3 and Lambda: examples/serverless_lancedb_with_s3_and_lambda.md
-              - Serverless QA Bot with Modal: examples/serverless_qa_bot_with_modal_and_langchain.md
+              - Build From Scratch: examples/python_examples/build_from_scratch.md
+              - Multimodal: examples/python_examples/multimodal.md
+              - Rag: examples/python_examples/rag.md
+              - Miscellaneous: 
+                  - YouTube Transcript Search: notebooks/youtube_transcript_search.ipynb
+                  - Documentation QA Bot using LangChain: notebooks/code_qa_bot.ipynb
+                  - Multimodal search using CLIP: notebooks/multimodal_search.ipynb
+                  - Serverless QA Bot with S3 and Lambda: examples/serverless_lancedb_with_s3_and_lambda.md
+                  - Serverless QA Bot with Modal: examples/serverless_qa_bot_with_modal_and_langchain.md
           - 👾 JavaScript:
               - Overview: examples/examples_js.md
               - Serverless Website Chatbot: examples/serverless_website_chatbot.md
@@ -221,14 +224,24 @@ nav:
       - PromptTools: integrations/prompttools.md
   - Examples:
       - examples/index.md
-      - YouTube Transcript Search: notebooks/youtube_transcript_search.ipynb
-      - Documentation QA Bot using LangChain: notebooks/code_qa_bot.ipynb
-      - Multimodal search using CLIP: notebooks/multimodal_search.ipynb
-      - Serverless QA Bot with S3 and Lambda: examples/serverless_lancedb_with_s3_and_lambda.md
-      - Serverless QA Bot with Modal: examples/serverless_qa_bot_with_modal_and_langchain.md
-      - YouTube Transcript Search (JS): examples/youtube_transcript_bot_with_nodejs.md
-      - Serverless Chatbot from any website: examples/serverless_website_chatbot.md
-      - TransformersJS Embedding Search: examples/transformerjs_embedding_search_nodejs.md
+      - 🐍 Python:
+          - Overview: examples/examples_python.md
+          - Build From Scratch: examples/python_examples/build_from_scratch.md
+          - Multimodal: examples/python_examples/multimodal.md
+          - Rag: examples/python_examples/rag.md
+          - Miscellaneous: 
+              - YouTube Transcript Search: notebooks/youtube_transcript_search.ipynb
+              - Documentation QA Bot using LangChain: notebooks/code_qa_bot.ipynb
+              - Multimodal search using CLIP: notebooks/multimodal_search.ipynb
+              - Serverless QA Bot with S3 and Lambda: examples/serverless_lancedb_with_s3_and_lambda.md
+              - Serverless QA Bot with Modal: examples/serverless_qa_bot_with_modal_and_langchain.md
+      - 👾 JavaScript:
+          - Overview: examples/examples_js.md
+          - Serverless Website Chatbot: examples/serverless_website_chatbot.md
+          - YouTube Transcript Search: examples/youtube_transcript_bot_with_nodejs.md
+          - TransformersJS Embedding Search: examples/transformerjs_embedding_search_nodejs.md
+      - 🦀 Rust:
+          - Overview: examples/examples_rust.md
   - API reference:
       - Overview: api_reference.md
       - Python: python/python.md

docs/src/assets/colab.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" width="117" height="20"><linearGradient id="b" x2="0" y2="100%"><stop offset="0" stop-color="#bbb" stop-opacity=".1"/><stop offset="1" stop-opacity=".1"/></linearGradient><clipPath id="a"><rect width="117" height="20" rx="3" fill="#fff"/></clipPath><g clip-path="url(#a)"><path fill="#555" d="M0 0h30v20H0z"/><path fill="#007ec6" d="M30 0h87v20H30z"/><path fill="url(#b)" d="M0 0h117v20H0z"/></g><g fill="#fff" text-anchor="middle" font-family="DejaVu Sans,Verdana,Geneva,sans-serif" font-size="110"><svg x="4px" y="0px" width="22px" height="20px" viewBox="-2 0 28 24" style="background-color: #fff;border-radius: 1px;"><path style="fill:#e8710a;" d="M1.977,16.77c-2.667-2.277-2.605-7.079,0-9.357C2.919,8.057,3.522,9.075,4.49,9.691c-1.152,1.6-1.146,3.201-0.004,4.803C3.522,15.111,2.918,16.126,1.977,16.77z"/><path style="fill:#f9ab00;" d="M12.257,17.114c-1.767-1.633-2.485-3.658-2.118-6.02c0.451-2.91,2.139-4.893,4.946-5.678c2.565-0.718,4.964-0.217,6.878,1.819c-0.884,0.743-1.707,1.547-2.434,2.446C18.488,8.827,17.319,8.435,16,8.856c-2.404,0.767-3.046,3.241-1.494,5.644c-0.241,0.275-0.493,0.541-0.721,0.826C13.295,15.939,12.511,16.3,12.257,17.114z"/><path style="fill:#e8710a;" d="M19.529,9.682c0.727-0.899,1.55-1.703,2.434-2.446c2.703,2.783,2.701,7.031-0.005,9.764c-2.648,2.674-6.936,2.725-9.701,0.115c0.254-0.814,1.038-1.175,1.528-1.788c0.228-0.285,0.48-0.552,0.721-0.826c1.053,0.916,2.254,1.268,3.6,0.83C20.502,14.551,21.151,11.927,19.529,9.682z"/><path style="fill:#f9ab00;" d="M4.49,9.691C3.522,9.075,2.919,8.057,1.977,7.413c2.209-2.398,5.721-2.942,8.476-1.355c0.555,0.32,0.719,0.606,0.285,1.128c-0.157,0.188-0.258,0.422-0.391,0.631c-0.299,0.47-0.509,1.067-0.929,1.371C8.933,9.539,8.523,8.847,8.021,8.746C6.673,8.475,5.509,8.787,4.49,9.691z"/><path style="fill:#f9ab00;" d="M1.977,16.77c0.941-0.644,1.545-1.659,2.509-2.277c1.373,1.152,2.85,1.433,4.45,0.499c0.332-0.194,0.503-0.088,0.673,0.19c0.386,0.635,0.753,1.285,1.181,1.89c0.34,0.48,0.222,0.715-0.253,1.006C7.84,19.73,4.205,19.188,1.977,16.77z"/></svg><text x="245" y="140" transform="scale(.1)" textLength="30"> </text><text x="725" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="770">Open in Colab</text><text x="725" y="140" transform="scale(.1)" textLength="770">Open in Colab</text></g> </svg>

docs/src/assets/ghost.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" width="88.25" height="28" role="img" aria-label="GHOST"><title>GHOST</title><g shape-rendering="crispEdges"><rect width="88.25" height="28" fill="#000"/></g><g fill="#fff" text-anchor="middle" font-family="Verdana,Geneva,DejaVu Sans,sans-serif" text-rendering="geometricPrecision" font-size="100"><image x="9" y="7" width="14" height="14" xlink:href="data:image/svg+xml;base64,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"/><text transform="scale(.1)" x="541.25" y="175" textLength="442.5" fill="#fff" font-weight="bold">GHOST</text></g></svg>

docs/src/assets/github.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" width="95.5" height="28" role="img" aria-label="GITHUB"><title>GITHUB</title><g shape-rendering="crispEdges"><rect width="95.5" height="28" fill="#121011"/></g><g fill="#fff" text-anchor="middle" font-family="Verdana,Geneva,DejaVu Sans,sans-serif" text-rendering="geometricPrecision" font-size="100"><image x="9" y="7" width="14" height="14" xlink:href="data:image/svg+xml;base64,PHN2ZyBmaWxsPSJ3aGl0ZSIgcm9sZT0iaW1nIiB2aWV3Qm94PSIwIDAgMjQgMjQiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyI+PHRpdGxlPkdpdEh1YjwvdGl0bGU+PHBhdGggZD0iTTEyIC4yOTdjLTYuNjMgMC0xMiA1LjM3My0xMiAxMiAwIDUuMzAzIDMuNDM4IDkuOCA4LjIwNSAxMS4zODUuNi4xMTMuODItLjI1OC44Mi0uNTc3IDAtLjI4NS0uMDEtMS4wNC0uMDE1LTIuMDQtMy4zMzguNzI0LTQuMDQyLTEuNjEtNC4wNDItMS42MUM0LjQyMiAxOC4wNyAzLjYzMyAxNy43IDMuNjMzIDE3LjdjLTEuMDg3LS43NDQuMDg0LS43MjkuMDg0LS43MjkgMS4yMDUuMDg0IDEuODM4IDEuMjM2IDEuODM4IDEuMjM2IDEuMDcgMS44MzUgMi44MDkgMS4zMDUgMy40OTUuOTk4LjEwOC0uNzc2LjQxNy0xLjMwNS43Ni0xLjYwNS0yLjY2NS0uMy01LjQ2Ni0xLjMzMi01LjQ2Ni01LjkzIDAtMS4zMS40NjUtMi4zOCAxLjIzNS0zLjIyLS4xMzUtLjMwMy0uNTQtMS41MjMuMTA1LTMuMTc2IDAgMCAxLjAwNS0uMzIyIDMuMyAxLjIzLjk2LS4yNjcgMS45OC0uMzk5IDMtLjQwNSAxLjAyLjAwNiAyLjA0LjEzOCAzIC40MDUgMi4yOC0xLjU1MiAzLjI4NS0xLjIzIDMuMjg1LTEuMjMuNjQ1IDEuNjUzLjI0IDIuODczLjEyIDMuMTc2Ljc2NS44NCAxLjIzIDEuOTEgMS4yMyAzLjIyIDAgNC42MS0yLjgwNSA1LjYyNS01LjQ3NSA1LjkyLjQyLjM2LjgxIDEuMDk2LjgxIDIuMjIgMCAxLjYwNi0uMDE1IDIuODk2LS4wMTUgMy4yODYgMCAuMzE1LjIxLjY5LjgyNS41N0MyMC41NjUgMjIuMDkyIDI0IDE3LjU5MiAyNCAxMi4yOTdjMC02LjYyNy01LjM3My0xMi0xMi0xMiIvPjwvc3ZnPg=="/><text transform="scale(.1)" x="577.5" y="175" textLength="515" fill="#fff" font-weight="bold">GITHUB</text></g></svg>

docs/src/assets/python.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" width="97.5" height="28" role="img" aria-label="PYTHON"><title>PYTHON</title><g shape-rendering="crispEdges"><rect width="97.5" height="28" fill="#3670a0"/></g><g fill="#fff" text-anchor="middle" font-family="Verdana,Geneva,DejaVu Sans,sans-serif" text-rendering="geometricPrecision" font-size="100"><image x="9" y="7" width="14" height="14" xlink:href="data:image/svg+xml;base64,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"/><text transform="scale(.1)" x="587.5" y="175" textLength="535" fill="#fff" font-weight="bold">PYTHON</text></g></svg>

docs/src/embeddings/custom_embedding_function.md

@@ -15,198 +15,226 @@ There is another optional layer of abstraction available: `TextEmbeddingFunction
 
 Let's implement `SentenceTransformerEmbeddings` class. All you need to do is implement the `generate_embeddings()` and `ndims` function to handle the input types you expect and register the class in the global `EmbeddingFunctionRegistry`
 
-```python
-from lancedb.embeddings import register
-from lancedb.util import attempt_import_or_raise
 
-@register("sentence-transformers")
-class SentenceTransformerEmbeddings(TextEmbeddingFunction):
-    name: str = "all-MiniLM-L6-v2"
-    # set more default instance vars like device, etc.
+=== "Python"
 
-    def __init__(self, **kwargs):
-        super().__init__(**kwargs)
-        self._ndims = None
-    
-    def generate_embeddings(self, texts):
-        return self._embedding_model().encode(list(texts), ...).tolist()
+    ```python
+    from lancedb.embeddings import register
+    from lancedb.util import attempt_import_or_raise
 
-    def ndims(self):
-        if self._ndims is None:
-            self._ndims = len(self.generate_embeddings("foo")[0])
-        return self._ndims
+    @register("sentence-transformers")
+    class SentenceTransformerEmbeddings(TextEmbeddingFunction):
+        name: str = "all-MiniLM-L6-v2"
+        # set more default instance vars like device, etc.
 
-    @cached(cache={}) 
-    def _embedding_model(self):
-        return sentence_transformers.SentenceTransformer(name)
-```
+        def __init__(self, **kwargs):
+            super().__init__(**kwargs)
+            self._ndims = None
 
-This is a stripped down version of our implementation of `SentenceTransformerEmbeddings` that removes certain optimizations and defaul settings.
+        def generate_embeddings(self, texts):
+            return self._embedding_model().encode(list(texts), ...).tolist()
+
+        def ndims(self):
+            if self._ndims is None:
+                self._ndims = len(self.generate_embeddings("foo")[0])
+            return self._ndims
+
+        @cached(cache={})
+        def _embedding_model(self):
+            return sentence_transformers.SentenceTransformer(name)
+    ```
+
+=== "TypeScript"
+
+    ```ts
+    --8<--- "nodejs/examples/custom_embedding_function.ts:imports"
+
+    --8<--- "nodejs/examples/custom_embedding_function.ts:embedding_impl"
+    ```
+
+
+This is a stripped down version of our implementation of `SentenceTransformerEmbeddings` that removes certain optimizations and default settings.
 
 Now you can use this embedding function to create your table schema and that's it! you can then ingest data and run queries without manually vectorizing the inputs.
 
-```python
-from lancedb.pydantic import LanceModel, Vector
+=== "Python"
 
-registry = EmbeddingFunctionRegistry.get_instance()
-stransformer = registry.get("sentence-transformers").create()
+    ```python
+    from lancedb.pydantic import LanceModel, Vector
 
-class TextModelSchema(LanceModel):
-    vector: Vector(stransformer.ndims) = stransformer.VectorField()
-    text: str = stransformer.SourceField()
+    registry = EmbeddingFunctionRegistry.get_instance()
+    stransformer = registry.get("sentence-transformers").create()
 
-tbl = db.create_table("table", schema=TextModelSchema)
+    class TextModelSchema(LanceModel):
+        vector: Vector(stransformer.ndims) = stransformer.VectorField()
+        text: str = stransformer.SourceField()
 
-tbl.add(pd.DataFrame({"text": ["halo", "world"]}))
-result = tbl.search("world").limit(5)
-```
+    tbl = db.create_table("table", schema=TextModelSchema)
 
-NOTE:
+    tbl.add(pd.DataFrame({"text": ["halo", "world"]}))
+    result = tbl.search("world").limit(5)
+    ```
 
-You can always implement the `EmbeddingFunction` interface directly if you want or need to, `TextEmbeddingFunction` just makes it much simpler and faster for you to do so, by setting up the boiler plat for text-specific use case
+=== "TypeScript"
+
+    ```ts
+    --8<--- "nodejs/examples/custom_embedding_function.ts:call_custom_function"
+    ```
+
+!!! note
+
+    You can always implement the `EmbeddingFunction` interface directly if you want or need to, `TextEmbeddingFunction` just makes it much simpler and faster for you to do so, by setting up the boiler plat for text-specific use case
 
 ## Multi-modal embedding function example
-You can also use the `EmbeddingFunction` interface to implement more complex workflows such as multi-modal embedding function support. LanceDB implements `OpenClipEmeddingFunction` class that suppports multi-modal seach. Here's the implementation that you can use as a reference to build your own multi-modal embedding functions.
+You can also use the `EmbeddingFunction` interface to implement more complex workflows such as multi-modal embedding function support.
 
-```python
-@register("open-clip")
-class OpenClipEmbeddings(EmbeddingFunction):
-    name: str = "ViT-B-32"
-    pretrained: str = "laion2b_s34b_b79k"
-    device: str = "cpu"
-    batch_size: int = 64
-    normalize: bool = True
-    _model = PrivateAttr()
-    _preprocess = PrivateAttr()
-    _tokenizer = PrivateAttr()
+=== "Python"
 
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        open_clip = attempt_import_or_raise("open_clip", "open-clip") # EmbeddingFunction util to import external libs and raise if not found
-        model, _, preprocess = open_clip.create_model_and_transforms(
-            self.name, pretrained=self.pretrained
-        )
-        model.to(self.device)
-        self._model, self._preprocess = model, preprocess
-        self._tokenizer = open_clip.get_tokenizer(self.name)
-        self._ndims = None
+    LanceDB implements `OpenClipEmeddingFunction` class that suppports multi-modal seach. Here's the implementation that you can use as a reference to build your own multi-modal embedding functions.
 
-    def ndims(self):
-        if self._ndims is None:
-            self._ndims = self.generate_text_embeddings("foo").shape[0]
-        return self._ndims
+    ```python
+    @register("open-clip")
+    class OpenClipEmbeddings(EmbeddingFunction):
+        name: str = "ViT-B-32"
+        pretrained: str = "laion2b_s34b_b79k"
+        device: str = "cpu"
+        batch_size: int = 64
+        normalize: bool = True
+        _model = PrivateAttr()
+        _preprocess = PrivateAttr()
+        _tokenizer = PrivateAttr()
 
-    def compute_query_embeddings(
-        self, query: Union[str, "PIL.Image.Image"], *args, **kwargs
-    ) -> List[np.ndarray]:
-        """
-        Compute the embeddings for a given user query
+        def __init__(self, *args, **kwargs):
+            super().__init__(*args, **kwargs)
+            open_clip = attempt_import_or_raise("open_clip", "open-clip") # EmbeddingFunction util to import external libs and raise if not found
+            model, _, preprocess = open_clip.create_model_and_transforms(
+                self.name, pretrained=self.pretrained
+            )
+            model.to(self.device)
+            self._model, self._preprocess = model, preprocess
+            self._tokenizer = open_clip.get_tokenizer(self.name)
+            self._ndims = None
 
-        Parameters
-        ----------
-        query : Union[str, PIL.Image.Image]
-            The query to embed. A query can be either text or an image.
-        """
-        if isinstance(query, str):
-            return [self.generate_text_embeddings(query)]
-        else:
+        def ndims(self):
+            if self._ndims is None:
+                self._ndims = self.generate_text_embeddings("foo").shape[0]
+            return self._ndims
+
+        def compute_query_embeddings(
+            self, query: Union[str, "PIL.Image.Image"], *args, **kwargs
+        ) -> List[np.ndarray]:
+            """
+            Compute the embeddings for a given user query
+
+            Parameters
+            ----------
+            query : Union[str, PIL.Image.Image]
+                The query to embed. A query can be either text or an image.
+            """
+            if isinstance(query, str):
+                return [self.generate_text_embeddings(query)]
+            else:
+                PIL = attempt_import_or_raise("PIL", "pillow")
+                if isinstance(query, PIL.Image.Image):
+                    return [self.generate_image_embedding(query)]
+                else:
+                    raise TypeError("OpenClip supports str or PIL Image as query")
+
+        def generate_text_embeddings(self, text: str) -> np.ndarray:
+            torch = attempt_import_or_raise("torch")
+            text = self.sanitize_input(text)
+            text = self._tokenizer(text)
+            text.to(self.device)
+            with torch.no_grad():
+                text_features = self._model.encode_text(text.to(self.device))
+                if self.normalize:
+                    text_features /= text_features.norm(dim=-1, keepdim=True)
+                return text_features.cpu().numpy().squeeze()
+
+        def sanitize_input(self, images: IMAGES) -> Union[List[bytes], np.ndarray]:
+            """
+            Sanitize the input to the embedding function.
+            """
+            if isinstance(images, (str, bytes)):
+                images = [images]
+            elif isinstance(images, pa.Array):
+                images = images.to_pylist()
+            elif isinstance(images, pa.ChunkedArray):
+                images = images.combine_chunks().to_pylist()
+            return images
+
+        def compute_source_embeddings(
+            self, images: IMAGES, *args, **kwargs
+        ) -> List[np.array]:
+            """
+            Get the embeddings for the given images
+            """
+            images = self.sanitize_input(images)
+            embeddings = []
+            for i in range(0, len(images), self.batch_size):
+                j = min(i + self.batch_size, len(images))
+                batch = images[i:j]
+                embeddings.extend(self._parallel_get(batch))
+            return embeddings
+
+        def _parallel_get(self, images: Union[List[str], List[bytes]]) -> List[np.ndarray]:
+            """
+            Issue concurrent requests to retrieve the image data
+            """
+            with concurrent.futures.ThreadPoolExecutor() as executor:
+                futures = [
+                    executor.submit(self.generate_image_embedding, image)
+                    for image in images
+                ]
+                return [future.result() for future in futures]
+
+        def generate_image_embedding(
+            self, image: Union[str, bytes, "PIL.Image.Image"]
+        ) -> np.ndarray:
+            """
+            Generate the embedding for a single image
+
+            Parameters
+            ----------
+            image : Union[str, bytes, PIL.Image.Image]
+                The image to embed. If the image is a str, it is treated as a uri.
+                If the image is bytes, it is treated as the raw image bytes.
+            """
+            torch = attempt_import_or_raise("torch")
+            # TODO handle retry and errors for https
+            image = self._to_pil(image)
+            image = self._preprocess(image).unsqueeze(0)
+            with torch.no_grad():
+                return self._encode_and_normalize_image(image)
+
+        def _to_pil(self, image: Union[str, bytes]):
             PIL = attempt_import_or_raise("PIL", "pillow")
-            if isinstance(query, PIL.Image.Image):
-                return [self.generate_image_embedding(query)]
-            else:
-                raise TypeError("OpenClip supports str or PIL Image as query")
+            if isinstance(image, bytes):
+                return PIL.Image.open(io.BytesIO(image))
+            if isinstance(image, PIL.Image.Image):
+                return image
+            elif isinstance(image, str):
+                parsed = urlparse.urlparse(image)
+                # TODO handle drive letter on windows.
+                if parsed.scheme == "file":
+                    return PIL.Image.open(parsed.path)
+                elif parsed.scheme == "":
+                    return PIL.Image.open(image if os.name == "nt" else parsed.path)
+                elif parsed.scheme.startswith("http"):
+                    return PIL.Image.open(io.BytesIO(url_retrieve(image)))
+                else:
+                    raise NotImplementedError("Only local and http(s) urls are supported")
 
-    def generate_text_embeddings(self, text: str) -> np.ndarray:
-        torch = attempt_import_or_raise("torch")
-        text = self.sanitize_input(text)
-        text = self._tokenizer(text)
-        text.to(self.device)
-        with torch.no_grad():
-            text_features = self._model.encode_text(text.to(self.device))
+        def _encode_and_normalize_image(self, image_tensor: "torch.Tensor"):
+            """
+            encode a single image tensor and optionally normalize the output
+            """
+            image_features = self._model.encode_image(image_tensor)
             if self.normalize:
-                text_features /= text_features.norm(dim=-1, keepdim=True)
-            return text_features.cpu().numpy().squeeze()
+                image_features /= image_features.norm(dim=-1, keepdim=True)
+            return image_features.cpu().numpy().squeeze()
+    ```
 
-    def sanitize_input(self, images: IMAGES) -> Union[List[bytes], np.ndarray]:
-        """
-        Sanitize the input to the embedding function.
-        """
-        if isinstance(images, (str, bytes)):
-            images = [images]
-        elif isinstance(images, pa.Array):
-            images = images.to_pylist()
-        elif isinstance(images, pa.ChunkedArray):
-            images = images.combine_chunks().to_pylist()
-        return images
+=== "TypeScript"
 
-    def compute_source_embeddings(
-        self, images: IMAGES, *args, **kwargs
-    ) -> List[np.array]:
-        """
-        Get the embeddings for the given images
-        """
-        images = self.sanitize_input(images)
-        embeddings = []
-        for i in range(0, len(images), self.batch_size):
-            j = min(i + self.batch_size, len(images))
-            batch = images[i:j]
-            embeddings.extend(self._parallel_get(batch))
-        return embeddings
-
-    def _parallel_get(self, images: Union[List[str], List[bytes]]) -> List[np.ndarray]:
-        """
-        Issue concurrent requests to retrieve the image data
-        """
-        with concurrent.futures.ThreadPoolExecutor() as executor:
-            futures = [
-                executor.submit(self.generate_image_embedding, image)
-                for image in images
-            ]
-            return [future.result() for future in futures]
-
-    def generate_image_embedding(
-        self, image: Union[str, bytes, "PIL.Image.Image"]
-    ) -> np.ndarray:
-        """
-        Generate the embedding for a single image
-
-        Parameters
-        ----------
-        image : Union[str, bytes, PIL.Image.Image]
-            The image to embed. If the image is a str, it is treated as a uri.
-            If the image is bytes, it is treated as the raw image bytes.
-        """
-        torch = attempt_import_or_raise("torch")
-        # TODO handle retry and errors for https
-        image = self._to_pil(image)
-        image = self._preprocess(image).unsqueeze(0)
-        with torch.no_grad():
-            return self._encode_and_normalize_image(image)
-
-    def _to_pil(self, image: Union[str, bytes]):
-        PIL = attempt_import_or_raise("PIL", "pillow")
-        if isinstance(image, bytes):
-            return PIL.Image.open(io.BytesIO(image))
-        if isinstance(image, PIL.Image.Image):
-            return image
-        elif isinstance(image, str):
-            parsed = urlparse.urlparse(image)
-            # TODO handle drive letter on windows.
-            if parsed.scheme == "file":
-                return PIL.Image.open(parsed.path)
-            elif parsed.scheme == "":
-                return PIL.Image.open(image if os.name == "nt" else parsed.path)
-            elif parsed.scheme.startswith("http"):
-                return PIL.Image.open(io.BytesIO(url_retrieve(image)))
-            else:
-                raise NotImplementedError("Only local and http(s) urls are supported")
-
-    def _encode_and_normalize_image(self, image_tensor: "torch.Tensor"):
-        """
-        encode a single image tensor and optionally normalize the output
-        """
-        image_features = self._model.encode_image(image_tensor)
-        if self.normalize:
-            image_features /= image_features.norm(dim=-1, keepdim=True)
-        return image_features.cpu().numpy().squeeze()
-```
+    Coming Soon! See this [issue](https://github.com/lancedb/lancedb/issues/1482) to track the status!

docs/src/embeddings/default_embedding_functions.md

@@ -518,6 +518,82 @@ tbl.add(df)
 rs = tbl.search("hello").limit(1).to_pandas()
 ```
 
+# IBM watsonx.ai Embeddings
+
+Generate text embeddings using IBM's watsonx.ai platform.
+
+## Supported Models
+
+You can find a list of supported models at [IBM watsonx.ai Documentation](https://dataplatform.cloud.ibm.com/docs/content/wsj/analyze-data/fm-models-embed.html?context=wx). The currently supported model names are:
+
+- `ibm/slate-125m-english-rtrvr`
+- `ibm/slate-30m-english-rtrvr`
+- `sentence-transformers/all-minilm-l12-v2`
+- `intfloat/multilingual-e5-large`
+
+## Parameters
+
+The following parameters can be passed to the `create` method:
+
+| Parameter  | Type     | Default Value                    | Description                                               |
+|------------|----------|----------------------------------|-----------------------------------------------------------|
+| name       | str      | "ibm/slate-125m-english-rtrvr"   | The model ID of the watsonx.ai model to use               |
+| api_key    | str      | None                             | Optional IBM Cloud API key (or set `WATSONX_API_KEY`)     |
+| project_id | str      | None                             | Optional watsonx project ID (or set `WATSONX_PROJECT_ID`) |
+| url        | str      | None                             | Optional custom URL for the watsonx.ai instance           |
+| params     | dict     | None                             | Optional additional parameters for the embedding model    |
+
+## Usage Example
+
+First, the watsonx.ai library is an optional dependency, so must be installed seperately:
+
+```
+pip install ibm-watsonx-ai
+```
+
+Optionally set environment variables (if not passing credentials to `create` directly):
+
+```sh
+export WATSONX_API_KEY="YOUR_WATSONX_API_KEY"
+export WATSONX_PROJECT_ID="YOUR_WATSONX_PROJECT_ID"
+```
+
+```python
+import os
+import lancedb
+from lancedb.pydantic import LanceModel, Vector
+from lancedb.embeddings import EmbeddingFunctionRegistry
+
+watsonx_embed = EmbeddingFunctionRegistry
+  .get_instance()
+  .get("watsonx")
+  .create(
+    name="ibm/slate-125m-english-rtrvr",
+    # Uncomment and set these if not using environment variables
+    # api_key="your_api_key_here",
+    # project_id="your_project_id_here",
+    # url="your_watsonx_url_here",
+    # params={...},
+  )
+
+class TextModel(LanceModel):
+    text: str = watsonx_embed.SourceField()
+    vector: Vector(watsonx_embed.ndims()) = watsonx_embed.VectorField()
+
+data = [
+    {"text": "hello world"},
+    {"text": "goodbye world"},
+]
+
+db = lancedb.connect("~/.lancedb")
+tbl = db.create_table("watsonx_test", schema=TextModel, mode="overwrite")
+
+tbl.add(data)
+
+rs = tbl.search("hello").limit(1).to_pandas()
+print(rs)
+```
+
 ## Multi-modal embedding functions
 Multi-modal embedding functions allow you to query your table using both images and text.
 
@@ -721,4 +797,4 @@ Usage Example:
     table.add(
       pd.DataFrame({"label": labels, "image_uri": uris, "image_bytes": image_bytes})
     )
-```
+```

docs/src/examples/index.md

@@ -10,7 +10,7 @@ LanceDB provides language APIs, allowing you to embed a database in your languag
 
 ## Applications powered by LanceDB
 
-| Project Name                                        | Description                                                                                                          | Screenshot                                |
-|-----------------------------------------------------|----------------------------------------------------------------------------------------------------------------------|-------------------------------------------|
-| [YOLOExplorer](https://github.com/lancedb/yoloexplorer) | Iterate on your YOLO / CV datasets using SQL, Vector semantic search, and more within seconds                  | ![YOLOExplorer](https://github.com/lancedb/vectordb-recipes/assets/15766192/ae513a29-8f15-4e0b-99a1-ccd8272b6131) |
-| [Website Chatbot (Deployable Vercel Template)](https://github.com/lancedb/lancedb-vercel-chatbot) | Create a chatbot from the sitemap of any website/docs of your choice. Built using vectorDB serverless native javascript package. | ![Chatbot](../assets/vercel-template.gif)    |
+| Project Name | Description |
+| --- | --- |
+| **Ultralytics Explorer 🚀**<br>[![Ultralytics](https://img.shields.io/badge/Ultralytics-Docs-green?labelColor=0f3bc4&style=flat-square&logo=https://cdn.prod.website-files.com/646dd1f1a3703e451ba81ecc/64994922cf2a6385a4bf4489_UltralyticsYOLO_mark_blue.svg&link=https://docs.ultralytics.com/datasets/explorer/)](https://docs.ultralytics.com/datasets/explorer/)<br>[![Open In Collab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ultralytics/ultralytics/blob/main/docs/en/datasets/explorer/explorer.ipynb) |  - 🔍 **Explore CV Datasets**: Semantic search, SQL queries, vector similarity, natural language.<br>- 🖥️ **GUI & Python API**: Seamless dataset interaction.<br>- ⚡ **Efficient & Scalable**: Leverages LanceDB for large datasets.<br>- 📊 **Detailed Analysis**: Easily analyze data patterns.<br>- 🌐 **Browser GUI Demo**: Create embeddings, search images, run queries. |
+| **Website Chatbot🤖**<br>[![GitHub](https://img.shields.io/badge/github-%23121011.svg?style=for-the-badge&logo=github&logoColor=white)](https://github.com/lancedb/lancedb-vercel-chatbot)<br>[![Deploy with Vercel](https://vercel.com/button)](https://vercel.com/new/clone?repository-url=https%3A%2F%2Fgithub.com%2Flancedb%2Flancedb-vercel-chatbot&amp;env=OPENAI_API_KEY&amp;envDescription=OpenAI%20API%20Key%20for%20chat%20completion.&amp;project-name=lancedb-vercel-chatbot&amp;repository-name=lancedb-vercel-chatbot&amp;demo-title=LanceDB%20Chatbot%20Demo&amp;demo-description=Demo%20website%20chatbot%20with%20LanceDB.&amp;demo-url=https%3A%2F%2Flancedb.vercel.app&amp;demo-image=https%3A%2F%2Fi.imgur.com%2FazVJtvr.png) | - 🌐 **Chatbot from Sitemap/Docs**: Create a chatbot using site or document context.<br>- 🚀 **Embed LanceDB in Next.js**: Lightweight, on-prem storage.<br>- 🧠 **AI-Powered Context Retrieval**: Efficiently access relevant data.<br>- 🔧 **Serverless & Native JS**: Seamless integration with Next.js.<br>- ⚡ **One-Click Deploy on Vercel**: Quick and easy setup.. |

docs/src/examples/python_examples/build_from_scratch.md

@@ -0,0 +1,13 @@
+# Build from Scratch with LanceDB 🚀
+
+Start building your GenAI applications from the ground up using LanceDB's efficient vector-based document retrieval capabilities! 📄
+
+#### Get Started in Minutes ⏱️
+
+These examples provide a solid foundation for building your own GenAI applications using LanceDB. Jump from idea to proof of concept quickly with applied examples. Get started and see what you can create! 💻
+
+| **Build From Scratch**                     | **Description**                                                                                              | **Links**                                                                                                                                                                                                                                                                                                                                                                                                                |
+|:-------------------------------------------|:-------------------------------------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Build RAG from Scratch🚀💻**             | 📝 Create a **Retrieval-Augmented Generation** (RAG) model from scratch using LanceDB.                       | [![GitHub](https://img.shields.io/badge/github-%23121011.svg?style=for-the-badge&logo=github&logoColor=white)](https://github.com/lancedb/vectordb-recipes/tree/main/tutorials/RAG-from-Scratch)<br>[![Open In Collab](https://colab.research.google.com/assets/colab-badge.svg)]()                                                                                                                                  |
+| **Local RAG from Scratch with Llama3🔥💡** | 🐫 Build a local RAG model using **Llama3** and **LanceDB** for fast and efficient text generation.          | [![GitHub](https://img.shields.io/badge/github-%23121011.svg?style=for-the-badge&logo=github&logoColor=white)](https://github.com/lancedb/vectordb-recipes/tree/main/tutorials/Local-RAG-from-Scratch)<br>[![Python](https://img.shields.io/badge/python-3670A0?style=for-the-badge&logo=python&logoColor=ffdd54)](https://github.com/lancedb/vectordb-recipes/blob/main/tutorials/Local-RAG-from-Scratch/rag.py)    |
+| **Multi-Head RAG from Scratch📚💻**        | 🤯 Develop a **Multi-Head RAG model** from scratch, enabling generation of text based on multiple documents. | [![GitHub](https://img.shields.io/badge/github-%23121011.svg?style=for-the-badge&logo=github&logoColor=white)](https://github.com/lancedb/vectordb-recipes/tree/main/tutorials/Multi-Head-RAG-from-Scratch)<br>[![Python](https://img.shields.io/badge/python-3670A0?style=for-the-badge&logo=python&logoColor=ffdd54)](https://github.com/lancedb/vectordb-recipes/tree/main/tutorials/Multi-Head-RAG-from-Scratch) |

docs/src/examples/python_examples/multimodal.md

@@ -0,0 +1,28 @@
+# Multimodal Search with LanceDB 🔍💡
+
+Experience the future of search with LanceDB's multimodal capabilities. Combine text and image queries to find the most relevant results in your corpus and unlock new possibilities! 🔓💡
+
+#### Explore the Future of Search 🚀
+
+Unlock the power of multimodal search with LanceDB, enabling efficient vector-based retrieval of text and image data! 📊💻
+
+
+
+| **Multimodal**  | **Description**  | **Links**  |
+|:----------------|:-----------------|:-----------|
+| **Multimodal CLIP: DiffusionDB 🌐💥**    | Revolutionize search with Multimodal CLIP and DiffusionDB, combining text and image understanding for a new dimension of discovery! 🔓 | [![GitHub](../../assets/github.svg)][Clip_diffusionDB_github] <br>[![Open In Collab](../../assets/colab.svg)][Clip_diffusionDB_colab] <br>[![Python](../../assets/python.svg)][Clip_diffusionDB_python] <br>[![Ghost](../../assets/ghost.svg)][Clip_diffusionDB_ghost] |
+| **Multimodal CLIP: Youtube Videos 📹👀** | Search Youtube videos using Multimodal CLIP, finding relevant content with ease and accuracy! 🎯                                                                                                                                                                | [![Github](../../assets/github.svg)][Clip_youtube_github]                                                                                                    <br>[![Open In Collab](../../assets/colab.svg)][Clip_youtube_colab]                                                                                           <br> [![Python](../../assets/python.svg)][Clip_youtube_python]         <br>[![Ghost](../../assets/ghost.svg)][Clip_youtube_python] |
+| **Multimodal Image + Text Search 📸🔍**               | Discover relevant documents and images with a single query, using LanceDB's multimodal search capabilities to bridge the gap between text and visuals! 🌉 | [![GitHub](../../assets/github.svg)](https://github.com/lancedb/vectordb-recipes/blob/main/examples/multimodal_search) <br>[![Open In Collab](../../assets/colab.svg)](https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/multimodal_search/main.ipynb) <br> [![Python](../../assets/python.svg)](https://github.com/lancedb/vectordb-recipes/blob/main/examples/multimodal_search/main.py)<br> [![Ghost](../../assets/ghost.svg)](https://blog.lancedb.com/multi-modal-ai-made-easy-with-lancedb-clip-5aaf8801c939/) |
+| **Cambrian-1: Vision-Centric Image Exploration 🔍👀** | Dive into vision-centric exploration of images with Cambrian-1, powered by LanceDB's multimodal search to uncover new insights! 🔎                        | [![GitHub](../../assets/github.svg)](https://www.kaggle.com/code/prasantdixit/cambrian-1-vision-centric-exploration-of-images/)<br>[![Open In Collab](../../assets/colab.svg)]() <br> [![Python](../../assets/python.svg)]() <br> [![Ghost](../../assets/ghost.svg)](https://blog.lancedb.com/cambrian-1-vision-centric-exploration/)                                                                                                                                                                                                                   |
+
+
+[Clip_diffusionDB_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/multimodal_clip_diffusiondb
+[Clip_diffusionDB_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/multimodal_clip_diffusiondb/main.ipynb
+[Clip_diffusionDB_python]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/multimodal_clip_diffusiondb/main.py
+[Clip_diffusionDB_ghost]: https://blog.lancedb.com/multi-modal-ai-made-easy-with-lancedb-clip-5aaf8801c939/
+
+
+[Clip_youtube_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/multimodal_video_search
+[Clip_youtube_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/multimodal_video_search/main.ipynb
+[Clip_youtube_python]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/multimodal_video_search/main.py
+[Clip_youtube_ghost]: https://blog.lancedb.com/multi-modal-ai-made-easy-with-lancedb-clip-5aaf8801c939/

docs/src/examples/python_examples/rag.md

@@ -0,0 +1,85 @@
+
+**🔍💡 RAG: Revolutionize Information Retrieval with LanceDB 🔓**
+====================================================================
+
+Unlock the full potential of Retrieval-Augmented Generation (RAG) with LanceDB, the ultimate solution for efficient vector-based information retrieval 📊. Input text queries and retrieve relevant documents with lightning-fast speed ⚡️ and accuracy ✅. Generate comprehensive answers by combining retrieved information, uncovering new insights 🔍 and connections.
+
+### Experience the Future of Search 🔄
+
+Experience the future of search with RAG, transforming information retrieval and answer generation. Apply RAG to various industries, streamlining processes 📈, saving time ⏰, and resources 💰. Stay ahead of the curve with innovative technology 🔝, powered by LanceDB. Discover the power of RAG with LanceDB and transform your industry with innovative solutions 💡.
+
+
+| **RAG**                                      | **Description**                                                                                                                                                  | **Links**                  |
+|----------------------------------------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------|----------------------------|                                                                                                                                                     
+| **RAG with Matryoshka Embeddings and LlamaIndex** 🪆🔗  | Utilize **Matryoshka embeddings** and **LlamaIndex** to improve the efficiency and accuracy of your RAG models. 📈✨ | [![Github](../../assets/github.svg)][matryoshka_github] <br>[![Open In Collab](../../assets/colab.svg)][matryoshka_colab]  |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **Improve RAG with Re-ranking** 📈🔄 | Enhance your RAG applications by implementing **re-ranking strategies** for more relevant document retrieval. 📚🔍 | [![Github](../../assets/github.svg)][rag_reranking_github] <br>[![Open In Collab](../../assets/colab.svg)][rag_reranking_colab] <br>[![Ghost](../../assets/ghost.svg)][rag_reranking_ghost] |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **Instruct-Multitask** 🧠🎯 |  Integrate the **Instruct Embedding Model** with LanceDB to streamline your embedding API, reducing redundant code and overhead. 🌐📊 | [![Github](../../assets/github.svg)][instruct_multitask_github] <br>[![Open In Collab](../../assets/colab.svg)][instruct_multitask_colab] <br>[![Python](../../assets/python.svg)][instruct_multitask_python] <br>[![Ghost](../../assets/ghost.svg)][instruct_multitask_ghost] |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **Improve RAG with HyDE** 🌌🔍 | Use **Hypothetical Document Embeddings** for efficient, accurate, and unsupervised dense retrieval. 📄🔍 | [![Github](../../assets/github.svg)][hyde_github] <br>[![Open In Collab](../../assets/colab.svg)][hyde_colab]<br>[![Ghost](../../assets/ghost.svg)][hyde_ghost]                 |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **Improve RAG with LOTR** 🧙‍♂️📜 | Enhance RAG with **Lord of the Retriever (LOTR)** to address 'Lost in the Middle' challenges, especially in medical data. 🌟📜 | [![Github](../../assets/github.svg)][lotr_github] <br>[![Open In Collab](../../assets/colab.svg)][lotr_colab] <br>[![Ghost](../../assets/ghost.svg)][lotr_ghost] |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **Advanced RAG: Parent Document Retriever** 📑🔗 | Use **Parent Document & Bigger Chunk Retriever** to maintain context and relevance when generating related content. 🎵📄 | [![Github](../../assets/github.svg)][parent_doc_retriever_github] <br>[![Open In Collab](../../assets/colab.svg)][parent_doc_retriever_colab] <br>[![Ghost](../../assets/ghost.svg)][parent_doc_retriever_ghost] |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **Corrective RAG with Langgraph** 🔧📊 | Enhance RAG reliability with **Corrective RAG (CRAG)** by self-reflecting and fact-checking for accurate and trustworthy results. ✅🔍 |[![Github](../../assets/github.svg)][corrective_rag_github] <br>[![Open In Collab](../../assets/colab.svg)][corrective_rag_colab] <br>[![Ghost](../../assets/ghost.svg)][corrective_rag_ghost] |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **Contextual Compression with RAG** 🗜️🧠 | Apply **contextual compression techniques** to condense large documents while retaining essential information. 📄🗜️                                               | [![Github](../../assets/github.svg)][compression_rag_github] <br>[![Open In Collab](../../assets/colab.svg)][compression_rag_colab] <br>[![Ghost](../../assets/ghost.svg)][compression_rag_ghost] |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **Improve RAG with FLARE** 🔥|  Enable users to ask questions directly to academic papers, focusing on ArXiv papers, with Forward-Looking Active REtrieval augmented generation.🚀🌟                                                                       | [![Github](../../assets/github.svg)][flare_github] <br>[![Open In Collab](../../assets/colab.svg)][flare_colab] <br>[![Ghost](../../assets/ghost.svg)][flare_ghost] |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **Query Expansion and Reranker** 🔍🔄 | Enhance RAG with query expansion using Large Language Models and advanced **reranking methods** like Cross Encoders, ColBERT v2, and FlashRank for improved document retrieval precision and recall 🔍📈                                                         | [![Github](../../assets/github.svg)][query_github] <br>[![Open In Collab](../../assets/colab.svg)][query_colab] |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **RAG Fusion** ⚡🌐 | Revolutionize search with RAG Fusion, utilizing the **RRF algorithm** to rerank documents based on user queries, and leveraging LanceDB and OPENAI Embeddings for efficient information retrieval ⚡🌐                                                                          | [![Github](../../assets/github.svg)][fusion_github] <br>[![Open In Collab](../../assets/colab.svg)][fusion_colab] |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
+| **Agentic RAG** 🤖📚 | Unlock autonomous information retrieval with **Agentic RAG**, a framework of **intelligent agents** that collaborate to synthesize, summarize, and compare data across sources, enabling proactive and informed decision-making 🤖📚                                                              | [![Github](../../assets/github.svg)][agentic_github] <br>[![Open In Collab](../../assets/colab.svg)][agentic_colab] |
+
+
+
+
+
+
+
+
+
+
+
+
+[matryoshka_github]: https://github.com/lancedb/vectordb-recipes/blob/main/tutorials/RAG-with_MatryoshkaEmbed-Llamaindex
+[matryoshka_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/tutorials/RAG-with_MatryoshkaEmbed-Llamaindex/RAG_with_MatryoshkaEmbedding_and_Llamaindex.ipynb
+
+[rag_reranking_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/RAG_Reranking
+[rag_reranking_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/RAG_Reranking/main.ipynb
+[rag_reranking_ghost]: https://blog.lancedb.com/simplest-method-to-improve-rag-pipeline-re-ranking-cf6eaec6d544
+
+
+[instruct_multitask_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/instruct-multitask
+[instruct_multitask_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/instruct-multitask/main.ipynb
+[instruct_multitask_python]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/instruct-multitask/main.py
+[instruct_multitask_ghost]: https://blog.lancedb.com/multitask-embedding-with-lancedb-be18ec397543
+
+[hyde_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/Advance-RAG-with-HyDE
+[hyde_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/Advance-RAG-with-HyDE/main.ipynb
+[hyde_ghost]: https://blog.lancedb.com/advanced-rag-precise-zero-shot-dense-retrieval-with-hyde-0946c54dfdcb
+
+[lotr_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/Advance_RAG_LOTR
+[lotr_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/Advance_RAG_LOTR/main.ipynb
+[lotr_ghost]: https://blog.lancedb.com/better-rag-with-lotr-lord-of-retriever-23c8336b9a35
+
+[parent_doc_retriever_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/parent_document_retriever
+[parent_doc_retriever_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/parent_document_retriever/main.ipynb
+[parent_doc_retriever_ghost]: https://blog.lancedb.com/modified-rag-parent-document-bigger-chunk-retriever-62b3d1e79bc6
+
+[corrective_rag_github]: https://github.com/lancedb/vectordb-recipes/blob/main/tutorials/Corrective-RAG-with_Langgraph
+[corrective_rag_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/tutorials/Corrective-RAG-with_Langgraph/CRAG_with_Langgraph.ipynb
+[corrective_rag_ghost]: https://blog.lancedb.com/implementing-corrective-rag-in-the-easiest-way-2/
+
+[compression_rag_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/Contextual-Compression-with-RAG
+[compression_rag_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/Contextual-Compression-with-RAG/main.ipynb
+[compression_rag_ghost]: https://blog.lancedb.com/enhance-rag-integrate-contextual-compression-and-filtering-for-precision-a29d4a810301/
+
+[flare_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/better-rag-FLAIR
+[flare_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/better-rag-FLAIR/main.ipynb
+[flare_ghost]: https://blog.lancedb.com/better-rag-with-active-retrieval-augmented-generation-flare-3b66646e2a9f/
+
+[query_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/QueryExpansion&Reranker
+[query_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/QueryExpansion&Reranker/main.ipynb
+
+
+[fusion_github]: https://github.com/lancedb/vectordb-recipes/blob/main/examples/RAG_Fusion
+[fusion_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/examples/RAG_Fusion/main.ipynb
+
+[agentic_github]: https://github.com/lancedb/vectordb-recipes/blob/main/tutorials/Agentic_RAG
+[agentic_colab]: https://colab.research.google.com/github/lancedb/vectordb-recipes/blob/main/tutorials/Agentic_RAG/main.ipynb
+
+

docs/test/requirements.txt

@@ -5,4 +5,5 @@ pylance
 duckdb
 --extra-index-url https://download.pytorch.org/whl/cpu
 torch
-polars
+polars>=0.19, <=1.3.0
+

node/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "vectordb",
-  "version": "0.7.2",
+  "version": "0.8.0",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "vectordb",
-      "version": "0.7.2",
+      "version": "0.8.0",
       "cpu": [
         "x64",
         "arm64"

node/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vectordb",
-  "version": "0.7.2",
+  "version": "0.9.0",
   "description": " Serverless, low-latency vector database for AI applications",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",

nodejs/Cargo.toml

@@ -20,7 +20,6 @@ napi = { version = "2.16.8", default-features = false, features = [
     "async",
 ] }
 napi-derive = "2.16.4"
-
 # Prevent dynamic linking of lzma, which comes from datafusion
 lzma-sys = { version = "*", features = ["static"] }
 

nodejs/__test__/registry.test.ts

@@ -1,3 +1,4 @@
+import * as apiArrow from "apache-arrow";
 // Copyright 2024 Lance Developers.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
@@ -69,7 +70,7 @@ describe.each([arrow13, arrow14, arrow15, arrow16, arrow17])(
           return 3;
         }
         embeddingDataType() {
-          return new arrow.Float32();
+          return new arrow.Float32() as apiArrow.Float;
         }
         async computeSourceEmbeddings(data: string[]) {
           return data.map(() => [1, 2, 3]);
@@ -82,7 +83,7 @@ describe.each([arrow13, arrow14, arrow15, arrow16, arrow17])(
 
       const schema = LanceSchema({
         id: new arrow.Int32(),
-        text: func.sourceField(new arrow.Utf8()),
+        text: func.sourceField(new arrow.Utf8() as apiArrow.DataType),
         vector: func.vectorField(),
       });
 
@@ -119,7 +120,7 @@ describe.each([arrow13, arrow14, arrow15, arrow16, arrow17])(
           return 3;
         }
         embeddingDataType() {
-          return new arrow.Float32();
+          return new arrow.Float32() as apiArrow.Float;
         }
         async computeSourceEmbeddings(data: string[]) {
           return data.map(() => [1, 2, 3]);
@@ -144,7 +145,7 @@ describe.each([arrow13, arrow14, arrow15, arrow16, arrow17])(
           return 3;
         }
         embeddingDataType() {
-          return new arrow.Float32();
+          return new arrow.Float32() as apiArrow.Float;
         }
         async computeSourceEmbeddings(data: string[]) {
           return data.map(() => [1, 2, 3]);
@@ -154,7 +155,7 @@ describe.each([arrow13, arrow14, arrow15, arrow16, arrow17])(
 
       const schema = LanceSchema({
         id: new arrow.Int32(),
-        text: func.sourceField(new arrow.Utf8()),
+        text: func.sourceField(new arrow.Utf8() as apiArrow.DataType),
         vector: func.vectorField(),
       });
       const expectedMetadata = new Map<string, string>([

nodejs/examples/custom_embedding_function.ts

@@ -0,0 +1,64 @@
+// --8<-- [start:imports]
+import * as lancedb from "@lancedb/lancedb";
+import {
+  LanceSchema,
+  TextEmbeddingFunction,
+  getRegistry,
+  register,
+} from "@lancedb/lancedb/embedding";
+import { pipeline } from "@xenova/transformers";
+// --8<-- [end:imports]
+
+// --8<-- [start:embedding_impl]
+@register("sentence-transformers")
+class SentenceTransformersEmbeddings extends TextEmbeddingFunction {
+  name = "Xenova/all-miniLM-L6-v2";
+  #ndims!: number;
+  extractor: any;
+
+  async init() {
+    this.extractor = await pipeline("feature-extraction", this.name);
+    this.#ndims = await this.generateEmbeddings(["hello"]).then(
+      (e) => e[0].length,
+    );
+  }
+
+  ndims() {
+    return this.#ndims;
+  }
+
+  toJSON() {
+    return {
+      name: this.name,
+    };
+  }
+  async generateEmbeddings(texts: string[]) {
+    const output = await this.extractor(texts, {
+      pooling: "mean",
+      normalize: true,
+    });
+    return output.tolist();
+  }
+}
+// -8<-- [end:embedding_impl]
+
+// --8<-- [start:call_custom_function]
+const registry = getRegistry();
+
+const sentenceTransformer = await registry
+  .get<SentenceTransformersEmbeddings>("sentence-transformers")!
+  .create();
+
+const schema = LanceSchema({
+  vector: sentenceTransformer.vectorField(),
+  text: sentenceTransformer.sourceField(),
+});
+
+const db = await lancedb.connect("/tmp/db");
+const table = await db.createEmptyTable("table", schema, { mode: "overwrite" });
+
+await table.add([{ text: "hello" }, { text: "world" }]);
+
+const results = await table.search("greeting").limit(1).toArray();
+console.log(results[0].text);
+// -8<-- [end:call_custom_function]

nodejs/lancedb/arrow.ts

@@ -103,50 +103,11 @@ export type IntoVector =
   | number[]
   | Promise<Float32Array | Float64Array | number[]>;
 
-export type FloatLike =
-  | import("apache-arrow-13").Float
-  | import("apache-arrow-14").Float
-  | import("apache-arrow-15").Float
-  | import("apache-arrow-16").Float
-  | import("apache-arrow-17").Float;
-export type DataTypeLike =
-  | import("apache-arrow-13").DataType
-  | import("apache-arrow-14").DataType
-  | import("apache-arrow-15").DataType
-  | import("apache-arrow-16").DataType
-  | import("apache-arrow-17").DataType;
-
 export function isArrowTable(value: object): value is TableLike {
   if (value instanceof ArrowTable) return true;
   return "schema" in value && "batches" in value;
 }
 
-export function isDataType(value: unknown): value is DataTypeLike {
-  return (
-    value instanceof DataType ||
-    DataType.isNull(value) ||
-    DataType.isInt(value) ||
-    DataType.isFloat(value) ||
-    DataType.isBinary(value) ||
-    DataType.isLargeBinary(value) ||
-    DataType.isUtf8(value) ||
-    DataType.isLargeUtf8(value) ||
-    DataType.isBool(value) ||
-    DataType.isDecimal(value) ||
-    DataType.isDate(value) ||
-    DataType.isTime(value) ||
-    DataType.isTimestamp(value) ||
-    DataType.isInterval(value) ||
-    DataType.isDuration(value) ||
-    DataType.isList(value) ||
-    DataType.isStruct(value) ||
-    DataType.isUnion(value) ||
-    DataType.isFixedSizeBinary(value) ||
-    DataType.isFixedSizeList(value) ||
-    DataType.isMap(value) ||
-    DataType.isDictionary(value)
-  );
-}
 export function isNull(value: unknown): value is Null {
   return value instanceof Null || DataType.isNull(value);
 }

nodejs/lancedb/connection.ts

@@ -44,10 +44,20 @@ export interface CreateTableOptions {
    * The available options are described at https://lancedb.github.io/lancedb/guides/storage/
    */
   storageOptions?: Record<string, string>;
+  /**
+   * The version of the data storage format to use.
+   *
+   * The default is `legacy`, which is Lance format v1.
+   * `stable` is the new format, which is Lance format v2.
+   */
+  dataStorageVersion?: string;
+
   /**
    * If true then data files will be written with the legacy format
    *
    * The default is true while the new format is in beta
+   *
+   * Deprecated.
    */
   useLegacyFormat?: boolean;
   schema?: SchemaLike;
@@ -247,12 +257,19 @@ export class LocalConnection extends Connection {
       throw new Error("data is required");
     }
     const { buf, mode } = await Table.parseTableData(data, options);
+    let dataStorageVersion = "legacy";
+    if (options?.dataStorageVersion !== undefined) {
+      dataStorageVersion = options.dataStorageVersion;
+    } else if (options?.useLegacyFormat !== undefined) {
+      dataStorageVersion = options.useLegacyFormat ? "legacy" : "stable";
+    }
+
     const innerTable = await this.inner.createTable(
       nameOrOptions,
       buf,
       mode,
       cleanseStorageOptions(options?.storageOptions),
-      options?.useLegacyFormat,
+      dataStorageVersion,
     );
 
     return new LocalTable(innerTable);
@@ -276,6 +293,13 @@ export class LocalConnection extends Connection {
       metadata = registry.getTableMetadata([embeddingFunction]);
     }
 
+    let dataStorageVersion = "legacy";
+    if (options?.dataStorageVersion !== undefined) {
+      dataStorageVersion = options.dataStorageVersion;
+    } else if (options?.useLegacyFormat !== undefined) {
+      dataStorageVersion = options.useLegacyFormat ? "legacy" : "stable";
+    }
+
     const table = makeEmptyTable(schema, metadata);
     const buf = await fromTableToBuffer(table);
     const innerTable = await this.inner.createEmptyTable(
@@ -283,7 +307,7 @@ export class LocalConnection extends Connection {
       buf,
       mode,
       cleanseStorageOptions(options?.storageOptions),
-      options?.useLegacyFormat,
+      dataStorageVersion,
     );
     return new LocalTable(innerTable);
   }

nodejs/lancedb/embedding/embedding_function.ts

@@ -15,13 +15,12 @@
 import "reflect-metadata";
 import {
   DataType,
-  DataTypeLike,
   Field,
   FixedSizeList,
+  Float,
   Float32,
-  FloatLike,
   type IntoVector,
-  isDataType,
+  Utf8,
   isFixedSizeList,
   isFloat,
   newVectorType,
@@ -93,11 +92,12 @@ export abstract class EmbeddingFunction<
    * @see {@link lancedb.LanceSchema}
    */
   sourceField(
-    optionsOrDatatype: Partial<FieldOptions> | DataTypeLike,
-  ): [DataTypeLike, Map<string, EmbeddingFunction>] {
-    let datatype = isDataType(optionsOrDatatype)
-      ? optionsOrDatatype
-      : optionsOrDatatype?.datatype;
+    optionsOrDatatype: Partial<FieldOptions> | DataType,
+  ): [DataType, Map<string, EmbeddingFunction>] {
+    let datatype =
+      "datatype" in optionsOrDatatype
+        ? optionsOrDatatype.datatype
+        : optionsOrDatatype;
     if (!datatype) {
       throw new Error("Datatype is required");
     }
@@ -123,15 +123,17 @@ export abstract class EmbeddingFunction<
     let dims: number | undefined = this.ndims();
 
     // `func.vectorField(new Float32())`
-    if (isDataType(optionsOrDatatype)) {
-      dtype = optionsOrDatatype;
+    if (optionsOrDatatype === undefined) {
+      dtype = new Float32();
+    } else if (!("datatype" in optionsOrDatatype)) {
+      dtype = sanitizeType(optionsOrDatatype);
     } else {
       // `func.vectorField({
       //  datatype: new Float32(),
       //  dims: 10
       // })`
       dims = dims ?? optionsOrDatatype?.dims;
-      dtype = optionsOrDatatype?.datatype;
+      dtype = sanitizeType(optionsOrDatatype?.datatype);
     }
 
     if (dtype !== undefined) {
@@ -173,7 +175,7 @@ export abstract class EmbeddingFunction<
   }
 
   /** The datatype of the embeddings */
-  abstract embeddingDataType(): FloatLike;
+  abstract embeddingDataType(): Float;
 
   /**
    * Creates a vector representation for the given values.
@@ -192,6 +194,38 @@ export abstract class EmbeddingFunction<
   }
 }
 
+/**
+ * an abstract class for implementing embedding functions that take text as input
+ */
+export abstract class TextEmbeddingFunction<
+  M extends FunctionOptions = FunctionOptions,
+> extends EmbeddingFunction<string, M> {
+  //** Generate the embeddings for the given texts */
+  abstract generateEmbeddings(
+    texts: string[],
+    // biome-ignore lint/suspicious/noExplicitAny: we don't know what the implementor will do
+    ...args: any[]
+  ): Promise<number[][] | Float32Array[] | Float64Array[]>;
+
+  async computeQueryEmbeddings(data: string): Promise<Awaited<IntoVector>> {
+    return this.generateEmbeddings([data]).then((data) => data[0]);
+  }
+
+  embeddingDataType(): Float {
+    return new Float32();
+  }
+
+  override sourceField(): [DataType, Map<string, EmbeddingFunction>] {
+    return super.sourceField(new Utf8());
+  }
+
+  computeSourceEmbeddings(
+    data: string[],
+  ): Promise<number[][] | Float32Array[] | Float64Array[]> {
+    return this.generateEmbeddings(data);
+  }
+}
+
 export interface FieldOptions<T extends DataType = DataType> {
   datatype: T;
   dims?: number;

nodejs/lancedb/embedding/index.ts

@@ -13,12 +13,11 @@
 // limitations under the License.
 
 import { Field, Schema } from "../arrow";
-import { isDataType } from "../arrow";
 import { sanitizeType } from "../sanitize";
 import { EmbeddingFunction } from "./embedding_function";
 import { EmbeddingFunctionConfig, getRegistry } from "./registry";
 
-export { EmbeddingFunction } from "./embedding_function";
+export { EmbeddingFunction, TextEmbeddingFunction } from "./embedding_function";
 
 // We need to explicitly export '*' so that the `register` decorator actually registers the class.
 export * from "./openai";
@@ -57,15 +56,15 @@ export function LanceSchema(
     Partial<EmbeddingFunctionConfig>
   >();
   Object.entries(fields).forEach(([key, value]) => {
-    if (isDataType(value)) {
-      arrowFields.push(new Field(key, sanitizeType(value), true));
-    } else {
+    if (Array.isArray(value)) {
       const [dtype, metadata] = value as [
         object,
         Map<string, EmbeddingFunction>,
       ];
       arrowFields.push(new Field(key, sanitizeType(dtype), true));
       parseEmbeddingFunctions(embeddingFunctions, key, metadata);
+    } else {
+      arrowFields.push(new Field(key, sanitizeType(value), true));
     }
   });
   const registry = getRegistry();

nodejs/lancedb/embedding/openai.ts

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 import type OpenAI from "openai";
-import { type EmbeddingCreateParams } from "openai/resources";
+import type { EmbeddingCreateParams } from "openai/resources/index";
 import { Float, Float32 } from "../arrow";
 import { EmbeddingFunction } from "./embedding_function";
 import { register } from "./registry";

nodejs/lancedb/index.ts

@@ -59,7 +59,7 @@ export {
 
 export { Index, IndexOptions, IvfPqOptions } from "./indices";
 
-export { Table, AddDataOptions, UpdateOptions } from "./table";
+export { Table, AddDataOptions, UpdateOptions, OptimizeOptions } from "./table";
 
 export * as embedding from "./embedding";
 

nodejs/lancedb/sanitize.ts

@@ -340,8 +340,14 @@ export function sanitizeType(typeLike: unknown): DataType<any> {
   if (typeof typeLike !== "object" || typeLike === null) {
     throw Error("Expected a Type but object was null/undefined");
   }
-  if (!("typeId" in typeLike) || !(typeof typeLike.typeId !== "function")) {
-    throw Error("Expected a Type to have a typeId function");
+  if (
+    !("typeId" in typeLike) ||
+    !(
+      typeof typeLike.typeId !== "function" ||
+      typeof typeLike.typeId !== "number"
+    )
+  ) {
+    throw Error("Expected a Type to have a typeId property");
   }
   let typeId: Type;
   if (typeof typeLike.typeId === "function") {

nodejs/npm/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.7.2",
+	"version": "0.9.0",
 	"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": "0.7.2",
+	"version": "0.9.0",
 	"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": "0.7.2",
+	"version": "0.9.0",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-gnu.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-gnu",
-	"version": "0.7.2",
+	"version": "0.9.0",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-gnu.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.7.2",
+	"version": "0.9.0",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",

nodejs/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@lancedb/lancedb",
-  "version": "0.7.2",
+  "version": "0.8.0",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@lancedb/lancedb",
-      "version": "0.7.2",
+      "version": "0.8.0",
       "cpu": [
         "x64",
         "arm64"

nodejs/package.json

@@ -10,7 +10,7 @@
     "vector database",
     "ann"
   ],
-  "version": "0.7.2",
+  "version": "0.9.0",
   "main": "dist/index.js",
   "exports": {
     ".": "./dist/index.js",

nodejs/src/connection.rs

@@ -13,13 +13,16 @@
 // limitations under the License.
 
 use std::collections::HashMap;
+use std::str::FromStr;
 
 use napi::bindgen_prelude::*;
 use napi_derive::*;
 
 use crate::table::Table;
 use crate::ConnectionOptions;
-use lancedb::connection::{ConnectBuilder, Connection as LanceDBConnection, CreateTableMode};
+use lancedb::connection::{
+    ConnectBuilder, Connection as LanceDBConnection, CreateTableMode, LanceFileVersion,
+};
 use lancedb::ipc::{ipc_file_to_batches, ipc_file_to_schema};
 
 #[napi]
@@ -120,7 +123,7 @@ impl Connection {
         buf: Buffer,
         mode: String,
         storage_options: Option<HashMap<String, String>>,
-        use_legacy_format: Option<bool>,
+        data_storage_options: Option<String>,
     ) -> napi::Result<Table> {
         let batches = ipc_file_to_batches(buf.to_vec())
             .map_err(|e| napi::Error::from_reason(format!("Failed to read IPC file: {}", e)))?;
@@ -131,8 +134,11 @@ impl Connection {
                 builder = builder.storage_option(key, value);
             }
         }
-        if let Some(use_legacy_format) = use_legacy_format {
-            builder = builder.use_legacy_format(use_legacy_format);
+        if let Some(data_storage_option) = data_storage_options.as_ref() {
+            builder = builder.data_storage_version(
+                LanceFileVersion::from_str(data_storage_option)
+                    .map_err(|e| napi::Error::from_reason(format!("{}", e)))?,
+            );
         }
         let tbl = builder
             .execute()
@@ -148,7 +154,7 @@ impl Connection {
         schema_buf: Buffer,
         mode: String,
         storage_options: Option<HashMap<String, String>>,
-        use_legacy_format: Option<bool>,
+        data_storage_options: Option<String>,
     ) -> napi::Result<Table> {
         let schema = ipc_file_to_schema(schema_buf.to_vec()).map_err(|e| {
             napi::Error::from_reason(format!("Failed to marshal schema from JS to Rust: {}", e))
@@ -163,8 +169,11 @@ impl Connection {
                 builder = builder.storage_option(key, value);
             }
         }
-        if let Some(use_legacy_format) = use_legacy_format {
-            builder = builder.use_legacy_format(use_legacy_format);
+        if let Some(data_storage_option) = data_storage_options.as_ref() {
+            builder = builder.data_storage_version(
+                LanceFileVersion::from_str(data_storage_option)
+                    .map_err(|e| napi::Error::from_reason(format!("{}", e)))?,
+            );
         }
         let tbl = builder
             .execute()

nodejs/src/table.rs

@@ -293,6 +293,7 @@ impl Table {
             .optimize(OptimizeAction::Prune {
                 older_than,
                 delete_unverified: None,
+                error_if_tagged_old_versions: None,
             })
             .await
             .default_error()?

python/.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.11.0"
+current_version = "0.12.0"
 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 = "0.11.0"
+version = "0.12.0"
 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==0.15.0",
+    "pylance==0.16.0",
     "ratelimiter~=1.0",
     "requests>=2.31.0",
     "retry>=0.9.2",
@@ -56,7 +56,7 @@ tests = [
     "pytest-asyncio",
     "duckdb",
     "pytz",
-    "polars>=0.19",
+    "polars>=0.19, <=1.3.0",
     "tantivy",
 ]
 dev = ["ruff", "pre-commit"]
@@ -76,6 +76,7 @@ embeddings = [
     "awscli>=1.29.57",
     "botocore>=1.31.57",
     "ollama",
+    "ibm-watsonx-ai>=1.1.2",
 ]
 azure = ["adlfs>=2024.2.0"]
 

python/python/lancedb/_lancedb.pyi

@@ -24,7 +24,7 @@ class Connection(object):
         mode: str,
         data: pa.RecordBatchReader,
         storage_options: Optional[Dict[str, str]] = None,
-        use_legacy_format: Optional[bool] = None,
+        data_storage_version: Optional[str] = None,
     ) -> Table: ...
     async def create_empty_table(
         self,
@@ -32,7 +32,7 @@ class Connection(object):
         mode: str,
         schema: pa.Schema,
         storage_options: Optional[Dict[str, str]] = None,
-        use_legacy_format: Optional[bool] = None,
+        data_storage_version: Optional[str] = None,
     ) -> Table: ...
 
 class Table:

python/python/lancedb/db.py

@@ -560,6 +560,7 @@ class AsyncConnection(object):
         fill_value: Optional[float] = None,
         storage_options: Optional[Dict[str, str]] = None,
         *,
+        data_storage_version: Optional[str] = None,
         use_legacy_format: Optional[bool] = None,
     ) -> AsyncTable:
         """Create an [AsyncTable][lancedb.table.AsyncTable] in the database.
@@ -603,9 +604,15 @@ class AsyncConnection(object):
             connection will be inherited by the table, but can be overridden here.
             See available options at
             https://lancedb.github.io/lancedb/guides/storage/
-        use_legacy_format: bool, optional, default True
+        data_storage_version: optional, str, default "legacy"
+            The version of the data storage format to use. Newer versions are more
+            efficient but require newer versions of lance to read.  The default is
+            "legacy" which will use the legacy v1 version.  See the user guide
+            for more details.
+        use_legacy_format: bool, optional, default True. (Deprecated)
             If True, use the legacy format for the table. If False, use the new format.
             The default is True while the new format is in beta.
+            This method is deprecated, use `data_storage_version` instead.
 
 
         Returns
@@ -765,13 +772,18 @@ class AsyncConnection(object):
         if mode == "create" and exist_ok:
             mode = "exist_ok"
 
+        if not data_storage_version:
+            data_storage_version = (
+                "legacy" if use_legacy_format is None or use_legacy_format else "stable"
+            )
+
         if data is None:
             new_table = await self._inner.create_empty_table(
                 name,
                 mode,
                 schema,
                 storage_options=storage_options,
-                use_legacy_format=use_legacy_format,
+                data_storage_version=data_storage_version,
             )
         else:
             data = data_to_reader(data, schema)
@@ -780,7 +792,7 @@ class AsyncConnection(object):
                 mode,
                 data,
                 storage_options=storage_options,
-                use_legacy_format=use_legacy_format,
+                data_storage_version=data_storage_version,
             )
 
         return AsyncTable(new_table)

python/python/lancedb/embeddings/__init__.py

@@ -26,3 +26,4 @@ from .transformers import TransformersEmbeddingFunction, ColbertEmbeddings
 from .imagebind import ImageBindEmbeddings
 from .utils import with_embeddings
 from .jinaai import JinaEmbeddings
+from .watsonx import WatsonxEmbeddings

python/python/lancedb/embeddings/watsonx.py

@@ -0,0 +1,111 @@
+#  Copyright (c) 2023. LanceDB Developers
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+import os
+from functools import cached_property
+from typing import List, Optional, Dict, Union
+
+from ..util import attempt_import_or_raise
+from .base import TextEmbeddingFunction
+from .registry import register
+
+import numpy as np
+
+DEFAULT_WATSONX_URL = "https://us-south.ml.cloud.ibm.com"
+
+MODELS_DIMS = {
+    "ibm/slate-125m-english-rtrvr": 768,
+    "ibm/slate-30m-english-rtrvr": 384,
+    "sentence-transformers/all-minilm-l12-v2": 384,
+    "intfloat/multilingual-e5-large": 1024,
+}
+
+
+@register("watsonx")
+class WatsonxEmbeddings(TextEmbeddingFunction):
+    """
+    API Docs:
+    ---------
+    https://cloud.ibm.com/apidocs/watsonx-ai#text-embeddings
+
+    Supported embedding models:
+    ---------------------------
+    https://dataplatform.cloud.ibm.com/docs/content/wsj/analyze-data/fm-models-embed.html?context=wx
+    """
+
+    name: str = "ibm/slate-125m-english-rtrvr"
+    api_key: Optional[str] = None
+    project_id: Optional[str] = None
+    url: Optional[str] = None
+    params: Optional[Dict] = None
+
+    @staticmethod
+    def model_names():
+        return [
+            "ibm/slate-125m-english-rtrvr",
+            "ibm/slate-30m-english-rtrvr",
+            "sentence-transformers/all-minilm-l12-v2",
+            "intfloat/multilingual-e5-large",
+        ]
+
+    def ndims(self):
+        return self._ndims
+
+    @cached_property
+    def _ndims(self):
+        if self.name not in MODELS_DIMS:
+            raise ValueError(f"Unknown model name {self.name}")
+        return MODELS_DIMS[self.name]
+
+    def generate_embeddings(
+        self,
+        texts: Union[List[str], np.ndarray],
+        *args,
+        **kwargs,
+    ) -> List[List[float]]:
+        return self._watsonx_client.embed_documents(
+            texts=list(texts),
+            *args,
+            **kwargs,
+        )
+
+    @cached_property
+    def _watsonx_client(self):
+        ibm_watsonx_ai = attempt_import_or_raise("ibm_watsonx_ai")
+        ibm_watsonx_ai_foundation_models = attempt_import_or_raise(
+            "ibm_watsonx_ai.foundation_models"
+        )
+
+        kwargs = {"model_id": self.name}
+        if self.params:
+            kwargs["params"] = self.params
+        if self.project_id:
+            kwargs["project_id"] = self.project_id
+        elif "WATSONX_PROJECT_ID" in os.environ:
+            kwargs["project_id"] = os.environ["WATSONX_PROJECT_ID"]
+        else:
+            raise ValueError("WATSONX_PROJECT_ID must be set or passed")
+
+        creds_kwargs = {}
+        if self.api_key:
+            creds_kwargs["api_key"] = self.api_key
+        elif "WATSONX_API_KEY" in os.environ:
+            creds_kwargs["api_key"] = os.environ["WATSONX_API_KEY"]
+        else:
+            raise ValueError("WATSONX_API_KEY must be set or passed")
+        if self.url:
+            creds_kwargs["url"] = self.url
+        else:
+            creds_kwargs["url"] = DEFAULT_WATSONX_URL
+        kwargs["credentials"] = ibm_watsonx_ai.Credentials(**creds_kwargs)
+
+        return ibm_watsonx_ai_foundation_models.Embeddings(**kwargs)

python/python/lancedb/remote/table.py

@@ -22,8 +22,9 @@ from lance import json_to_schema
 
 from lancedb.common import DATA, VEC, VECTOR_COLUMN_NAME
 from lancedb.merge import LanceMergeInsertBuilder
+from lancedb.embeddings import EmbeddingFunctionRegistry
 
-from ..query import LanceVectorQueryBuilder
+from ..query import LanceVectorQueryBuilder, LanceQueryBuilder
 from ..table import Query, Table, _sanitize_data
 from ..util import inf_vector_column_query, value_to_sql
 from .arrow import to_ipc_binary
@@ -58,6 +59,21 @@ class RemoteTable(Table):
         resp = self._conn._client.post(f"/v1/table/{self._name}/describe/")
         return resp["version"]
 
+    @cached_property
+    def embedding_functions(self) -> dict:
+        """
+        Get the embedding functions for the table
+
+        Returns
+        -------
+        funcs: dict
+            A mapping of the vector column to the embedding function
+            or empty dict if not configured.
+        """
+        return EmbeddingFunctionRegistry.get_instance().parse_functions(
+            self.schema.metadata
+        )
+
     def to_arrow(self) -> pa.Table:
         """to_arrow() is not yet supported on LanceDB cloud."""
         raise NotImplementedError("to_arrow() is not yet supported on LanceDB cloud.")
@@ -213,7 +229,7 @@ class RemoteTable(Table):
         data, _ = _sanitize_data(
             data,
             self.schema,
-            metadata=None,
+            metadata=self.schema.metadata,
             on_bad_vectors=on_bad_vectors,
             fill_value=fill_value,
         )
@@ -293,6 +309,7 @@ class RemoteTable(Table):
         """
         if vector_column_name is None:
             vector_column_name = inf_vector_column_query(self.schema)
+        query = LanceQueryBuilder._query_to_vector(self, query, vector_column_name)
         return LanceVectorQueryBuilder(self, query, vector_column_name)
 
     def _execute_query(
@@ -336,7 +353,7 @@ class RemoteTable(Table):
 
         See [`Table.merge_insert`][lancedb.table.Table.merge_insert] for more details.
         """
-        super().merge_insert(on)
+        return super().merge_insert(on)
 
     def _do_merge(
         self,

python/python/lancedb/rerankers/base.py

@@ -1,9 +1,13 @@
 from abc import ABC, abstractmethod
 from packaging.version import Version
+from typing import Union, List, TYPE_CHECKING
 
 import numpy as np
 import pyarrow as pa
 
+if TYPE_CHECKING:
+    from ..table import LanceVectorQueryBuilder
+
 ARROW_VERSION = Version(pa.__version__)
 
 
@@ -130,12 +134,94 @@ class Reranker(ABC):
         combined = pa.concat_tables(
             [vector_results, fts_results], **self._concat_tables_args
         )
-        row_id = combined.column("_rowid")
 
         # deduplicate
-        mask = np.full((combined.shape[0]), False)
-        _, mask_indices = np.unique(np.array(row_id), return_index=True)
-        mask[mask_indices] = True
-        combined = combined.filter(mask=mask)
+        combined = self._deduplicate(combined)
 
         return combined
+
+    def rerank_multivector(
+        self,
+        vector_results: Union[List[pa.Table], List["LanceVectorQueryBuilder"]],
+        query: Union[str, None],  # Some rerankers might not need the query
+        deduplicate: bool = False,
+    ):
+        """
+        This is a rerank function that receives the results from multiple
+        vector searches. For example, this can be used to combine the
+        results of two vector searches with different embeddings.
+
+        Parameters
+        ----------
+        vector_results : List[pa.Table] or List[LanceVectorQueryBuilder]
+            The results from the vector search. Either accepts the query builder
+            if the results haven't been executed yet or the results in arrow format.
+        query : str or None,
+            The input query. Some rerankers might not need the query to rerank.
+            In that case, it can be set to None explicitly. This is inteded to
+            be handled by the reranker implementations.
+        deduplicate : bool, optional
+            Whether to deduplicate the results based on the `_rowid` column,
+            by default False. Requires `_rowid` to be present in the results.
+
+        Returns
+        -------
+        pa.Table
+            The reranked results
+        """
+        vector_results = (
+            [vector_results] if not isinstance(vector_results, list) else vector_results
+        )
+
+        # Make sure all elements are of the same type
+        if not all(isinstance(v, type(vector_results[0])) for v in vector_results):
+            raise ValueError(
+                "All elements in vector_results should be of the same type"
+            )
+
+        # avoids circular import
+        if type(vector_results[0]).__name__ == "LanceVectorQueryBuilder":
+            vector_results = [result.to_arrow() for result in vector_results]
+        elif not isinstance(vector_results[0], pa.Table):
+            raise ValueError(
+                "vector_results should be a list of pa.Table or LanceVectorQueryBuilder"
+            )
+
+        combined = pa.concat_tables(vector_results, **self._concat_tables_args)
+
+        reranked = self.rerank_vector(query, combined)
+
+        # TODO: Allow custom deduplicators here.
+        # currently, this'll just keep the first instance.
+        if deduplicate:
+            if "_rowid" not in combined.column_names:
+                raise ValueError(
+                    "'_rowid' is required for deduplication. \
+                    add _rowid to search results like this: \
+                    `search().with_row_id(True)`"
+                )
+            reranked = self._deduplicate(reranked)
+
+        return reranked
+
+    def _deduplicate(self, table: pa.Table):
+        """
+        Deduplicate the table based on the `_rowid` column.
+        """
+        row_id = table.column("_rowid")
+
+        # deduplicate
+        mask = np.full((table.shape[0]), False)
+        _, mask_indices = np.unique(np.array(row_id), return_index=True)
+        mask[mask_indices] = True
+        deduped_table = table.filter(mask=mask)
+
+        return deduped_table
+
+    def _keep_relevance_score(self, combined_results: pa.Table):
+        if self.score == "relevance":
+            if "score" in combined_results.column_names:
+                combined_results = combined_results.drop_columns(["score"])
+            if "_distance" in combined_results.column_names:
+                combined_results = combined_results.drop_columns(["_distance"])
+        return combined_results

python/python/lancedb/rerankers/cohere.py

@@ -88,7 +88,7 @@ class CohereReranker(Reranker):
         combined_results = self.merge_results(vector_results, fts_results)
         combined_results = self._rerank(combined_results, query)
         if self.score == "relevance":
-            combined_results = combined_results.drop_columns(["score", "_distance"])
+            combined_results = self._keep_relevance_score(combined_results)
         elif self.score == "all":
             raise NotImplementedError(
                 "return_score='all' not implemented for cohere reranker"

python/python/lancedb/rerankers/colbert.py

@@ -73,7 +73,7 @@ class ColbertReranker(Reranker):
         combined_results = self.merge_results(vector_results, fts_results)
         combined_results = self._rerank(combined_results, query)
         if self.score == "relevance":
-            combined_results = combined_results.drop_columns(["score", "_distance"])
+            combined_results = self._keep_relevance_score(combined_results)
         elif self.score == "all":
             raise NotImplementedError(
                 "OpenAI Reranker does not support score='all' yet"

python/python/lancedb/rerankers/cross_encoder.py

@@ -66,7 +66,7 @@ class CrossEncoderReranker(Reranker):
         combined_results = self._rerank(combined_results, query)
         # sort the results by _score
         if self.score == "relevance":
-            combined_results = combined_results.drop_columns(["score", "_distance"])
+            combined_results = self._keep_relevance_score(combined_results)
         elif self.score == "all":
             raise NotImplementedError(
                 "return_score='all' not implemented for CrossEncoderReranker"

python/python/lancedb/rerankers/jinaai.py

@@ -92,7 +92,7 @@ class JinaReranker(Reranker):
         combined_results = self.merge_results(vector_results, fts_results)
         combined_results = self._rerank(combined_results, query)
         if self.score == "relevance":
-            combined_results = combined_results.drop_columns(["score", "_distance"])
+            combined_results = self._keep_relevance_score(combined_results)
         elif self.score == "all":
             raise NotImplementedError(
                 "return_score='all' not implemented for JinaReranker"

python/python/lancedb/rerankers/linear_combination.py

@@ -103,7 +103,7 @@ class LinearCombinationReranker(Reranker):
             [("_relevance_score", "descending")]
         )
         if self.score == "relevance":
-            tbl = tbl.drop_columns(["score", "_distance"])
+            tbl = self._keep_relevance_score(tbl)
         return tbl
 
     def _combine_score(self, score1, score2):

python/python/lancedb/rerankers/openai.py

@@ -84,7 +84,7 @@ class OpenaiReranker(Reranker):
         combined_results = self.merge_results(vector_results, fts_results)
         combined_results = self._rerank(combined_results, query)
         if self.score == "relevance":
-            combined_results = combined_results.drop_columns(["score", "_distance"])
+            combined_results = self._keep_relevance_score(combined_results)
         elif self.score == "all":
             raise NotImplementedError(
                 "OpenAI Reranker does not support score='all' yet"

python/python/lancedb/rerankers/rrf.py

@@ -1,8 +1,12 @@
+from typing import Union, List, TYPE_CHECKING
 import pyarrow as pa
 
 from collections import defaultdict
 from .base import Reranker
 
+if TYPE_CHECKING:
+    from ..table import LanceVectorQueryBuilder
+
 
 class RRFReranker(Reranker):
     """
@@ -55,6 +59,46 @@ class RRFReranker(Reranker):
         )
 
         if self.score == "relevance":
-            combined_results = combined_results.drop_columns(["score", "_distance"])
+            combined_results = self._keep_relevance_score(combined_results)
 
         return combined_results
+
+    def rerank_multivector(
+        self,
+        vector_results: Union[List[pa.Table], List["LanceVectorQueryBuilder"]],
+        query: str = None,
+        deduplicate: bool = True,  # noqa: F821 # TODO: automatically deduplicates
+    ):
+        """
+        Overridden method to rerank the results from multiple vector searches.
+        This leverages the RRF hybrid reranking algorithm to combine the
+        results from multiple vector searches as it doesn't support reranking
+        vector results individually.
+        """
+        # Make sure all elements are of the same type
+        if not all(isinstance(v, type(vector_results[0])) for v in vector_results):
+            raise ValueError(
+                "All elements in vector_results should be of the same type"
+            )
+
+        # avoid circular import
+        if type(vector_results[0]).__name__ == "LanceVectorQueryBuilder":
+            vector_results = [result.to_arrow() for result in vector_results]
+        elif not isinstance(vector_results[0], pa.Table):
+            raise ValueError(
+                "vector_results should be a list of pa.Table or LanceVectorQueryBuilder"
+            )
+
+        # _rowid is required for RRF reranking
+        if not all("_rowid" in result.column_names for result in vector_results):
+            raise ValueError(
+                "'_rowid' is required for deduplication. \
+                    add _rowid to search results like this: \
+                    `search().with_row_id(True)`"
+            )
+
+        combined = pa.concat_tables(vector_results, **self._concat_tables_args)
+        empty_table = pa.Table.from_arrays([], names=[])
+        reranked = self.rerank_hybrid(query, combined, empty_table)
+
+        return reranked

python/python/tests/test_embeddings_slow.py

@@ -417,3 +417,28 @@ def test_openai_embedding(tmp_path):
     tbl.add(df)
     assert len(tbl.to_pandas()["vector"][0]) == model.ndims()
     assert tbl.search("hello").limit(1).to_pandas()["text"][0] == "hello world"
+
+
+@pytest.mark.slow
+@pytest.mark.skipif(
+    os.environ.get("WATSONX_API_KEY") is None
+    or os.environ.get("WATSONX_PROJECT_ID") is None,
+    reason="WATSONX_API_KEY and WATSONX_PROJECT_ID not set",
+)
+def test_watsonx_embedding(tmp_path):
+    from lancedb.embeddings import WatsonxEmbeddings
+
+    for name in WatsonxEmbeddings.model_names():
+        model = get_registry().get("watsonx").create(max_retries=0, name=name)
+
+        class TextModel(LanceModel):
+            text: str = model.SourceField()
+            vector: Vector(model.ndims()) = model.VectorField()
+
+        db = lancedb.connect("~/.lancedb")
+        tbl = db.create_table("watsonx_test", schema=TextModel, mode="overwrite")
+        df = pd.DataFrame({"text": ["hello world", "goodbye world"]})
+
+        tbl.add(df)
+        assert len(tbl.to_pandas()["vector"][0]) == model.ndims()
+        assert tbl.search("hello").limit(1).to_pandas()["text"][0] == "hello world"

python/python/tests/test_index.py

@@ -42,6 +42,7 @@ async def test_create_scalar_index(some_table: AsyncTable):
     # Can recreate if replace=True
     await some_table.create_index("id", replace=True)
     indices = await some_table.list_indices()
+    assert str(indices) == '[Index(BTree, columns=["id"])]'
     assert len(indices) == 1
     assert indices[0].index_type == "BTree"
     assert indices[0].columns == ["id"]

python/python/tests/test_rerankers.py

@@ -1,4 +1,5 @@
 import os
+import random
 
 import lancedb
 import numpy as np
@@ -25,10 +26,13 @@ def get_test_table(tmp_path):
     db = lancedb.connect(tmp_path)
     # Create a LanceDB table schema with a vector and a text column
     emb = EmbeddingFunctionRegistry.get_instance().get("test")()
+    meta_emb = EmbeddingFunctionRegistry.get_instance().get("test")()
 
     class MyTable(LanceModel):
         text: str = emb.SourceField()
         vector: Vector(emb.ndims()) = emb.VectorField()
+        meta: str = meta_emb.SourceField()
+        meta_vector: Vector(meta_emb.ndims()) = meta_emb.VectorField()
 
     # Initialize the table using the schema
     table = LanceTable.create(
@@ -77,7 +81,12 @@ def get_test_table(tmp_path):
     ]
 
     # Add the phrases and vectors to the table
-    table.add([{"text": p} for p in phrases])
+    table.add(
+        [
+            {"text": p, "meta": phrases[random.randint(0, len(phrases) - 1)]}
+            for p in phrases
+        ]
+    )
 
     # Create a fts index
     table.create_fts_index("text")
@@ -88,12 +97,12 @@ def get_test_table(tmp_path):
 def _run_test_reranker(reranker, table, query, query_vector, schema):
     # Hybrid search setting
     result1 = (
-        table.search(query, query_type="hybrid")
+        table.search(query, query_type="hybrid", vector_column_name="vector")
         .rerank(normalize="score", reranker=reranker)
         .to_pydantic(schema)
     )
     result2 = (
-        table.search(query, query_type="hybrid")
+        table.search(query, query_type="hybrid", vector_column_name="vector")
         .rerank(reranker=reranker)
         .to_pydantic(schema)
     )
@@ -101,7 +110,7 @@ def _run_test_reranker(reranker, table, query, query_vector, schema):
 
     query_vector = table.to_pandas()["vector"][0]
     result = (
-        table.search((query_vector, query))
+        table.search((query_vector, query), vector_column_name="vector")
         .limit(30)
         .rerank(reranker=reranker)
         .to_arrow()
@@ -116,11 +125,16 @@ def _run_test_reranker(reranker, table, query, query_vector, schema):
     assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
 
     # Vector search setting
-    result = table.search(query).rerank(reranker=reranker).limit(30).to_arrow()
+    result = (
+        table.search(query, vector_column_name="vector")
+        .rerank(reranker=reranker)
+        .limit(30)
+        .to_arrow()
+    )
     assert len(result) == 30
     assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
     result_explicit = (
-        table.search(query_vector)
+        table.search(query_vector, vector_column_name="vector")
         .rerank(reranker=reranker, query_string=query)
         .limit(30)
         .to_arrow()
@@ -129,11 +143,13 @@ def _run_test_reranker(reranker, table, query, query_vector, schema):
     with pytest.raises(
         ValueError
     ):  # This raises an error because vector query is provided without reanking query
-        table.search(query_vector).rerank(reranker=reranker).limit(30).to_arrow()
+        table.search(query_vector, vector_column_name="vector").rerank(
+            reranker=reranker
+        ).limit(30).to_arrow()
 
     # FTS search setting
     result = (
-        table.search(query, query_type="fts")
+        table.search(query, query_type="fts", vector_column_name="vector")
         .rerank(reranker=reranker)
         .limit(30)
         .to_arrow()
@@ -141,22 +157,48 @@ def _run_test_reranker(reranker, table, query, query_vector, schema):
     assert len(result) > 0
     assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
 
+    # Multi-vector search setting
+    rs1 = table.search(query, vector_column_name="vector").limit(10).with_row_id(True)
+    rs2 = (
+        table.search(query, vector_column_name="meta_vector")
+        .limit(10)
+        .with_row_id(True)
+    )
+    result = reranker.rerank_multivector([rs1, rs2], query)
+    assert len(result) == 20
+    result_deduped = reranker.rerank_multivector(
+        [rs1, rs2, rs1], query, deduplicate=True
+    )
+    assert len(result_deduped) < 20
+    result_arrow = reranker.rerank_multivector([rs1.to_arrow(), rs2.to_arrow()], query)
+    assert len(result) == 20 and result == result_arrow
+
 
 def _run_test_hybrid_reranker(reranker, tmp_path):
     table, schema = get_test_table(tmp_path)
     # The default reranker
     result1 = (
-        table.search("Our father who art in heaven", query_type="hybrid")
+        table.search(
+            "Our father who art in heaven",
+            query_type="hybrid",
+            vector_column_name="vector",
+        )
         .rerank(normalize="score")
         .to_pydantic(schema)
     )
     result2 = (  # noqa
-        table.search("Our father who art in heaven.", query_type="hybrid")
+        table.search(
+            "Our father who art in heaven.",
+            query_type="hybrid",
+            vector_column_name="vector",
+        )
         .rerank(normalize="rank")
         .to_pydantic(schema)
     )
     result3 = table.search(
-        "Our father who art in heaven..", query_type="hybrid"
+        "Our father who art in heaven..",
+        query_type="hybrid",
+        vector_column_name="vector",
     ).to_pydantic(schema)
 
     assert result1 == result3  # 2 & 3 should be the same as they use score as score
@@ -164,7 +206,7 @@ def _run_test_hybrid_reranker(reranker, tmp_path):
     query = "Our father who art in heaven"
     query_vector = table.to_pandas()["vector"][0]
     result = (
-        table.search((query_vector, query))
+        table.search((query_vector, query), vector_column_name="vector")
         .limit(30)
         .rerank(normalize="score")
         .to_arrow()

python/python/tests/test_table.py

@@ -730,7 +730,7 @@ def test_create_scalar_index(db):
     indices = table.to_lance().list_indices()
     assert len(indices) == 1
     scalar_index = indices[0]
-    assert scalar_index["type"] == "Scalar"
+    assert scalar_index["type"] == "BTree"
 
     # Confirm that prefiltering still works with the scalar index column
     results = table.search().where("x = 'c'").to_arrow()
@@ -1034,6 +1034,12 @@ async def test_optimize(db_async: AsyncConnection):
         ],
     )
     stats = await table.optimize()
+    expected = (
+        "OptimizeStats(compaction=CompactionStats { fragments_removed: 2, "
+        "fragments_added: 1, files_removed: 2, files_added: 1 }, "
+        "prune=RemovalStats { bytes_removed: 0, old_versions_removed: 0 })"
+    )
+    assert str(stats) == expected
     assert stats.compaction.files_removed == 2
     assert stats.compaction.files_added == 1
     assert stats.compaction.fragments_added == 1

python/src/connection.rs

@@ -1,21 +1,10 @@
-// Copyright 2024 Lance Developers.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
-use std::{collections::HashMap, sync::Arc, time::Duration};
+use std::{collections::HashMap, str::FromStr, sync::Arc, time::Duration};
 
 use arrow::{datatypes::Schema, ffi_stream::ArrowArrayStreamReader, pyarrow::FromPyArrow};
-use lancedb::connection::{Connection as LanceConnection, CreateTableMode};
+use lancedb::connection::{Connection as LanceConnection, CreateTableMode, LanceFileVersion};
 use pyo3::{
     exceptions::{PyRuntimeError, PyValueError},
     pyclass, pyfunction, pymethods, Bound, PyAny, PyRef, PyResult, Python,
@@ -91,7 +80,7 @@ impl Connection {
         mode: &str,
         data: Bound<'_, PyAny>,
         storage_options: Option<HashMap<String, String>>,
-        use_legacy_format: Option<bool>,
+        data_storage_version: Option<String>,
     ) -> PyResult<Bound<'a, PyAny>> {
         let inner = self_.get_inner()?.clone();
 
@@ -104,8 +93,11 @@ impl Connection {
             builder = builder.storage_options(storage_options);
         }
 
-        if let Some(use_legacy_format) = use_legacy_format {
-            builder = builder.use_legacy_format(use_legacy_format);
+        if let Some(data_storage_version) = data_storage_version.as_ref() {
+            builder = builder.data_storage_version(
+                LanceFileVersion::from_str(data_storage_version)
+                    .map_err(|e| PyValueError::new_err(e.to_string()))?,
+            );
         }
 
         future_into_py(self_.py(), async move {
@@ -120,7 +112,7 @@ impl Connection {
         mode: &str,
         schema: Bound<'_, PyAny>,
         storage_options: Option<HashMap<String, String>>,
-        use_legacy_format: Option<bool>,
+        data_storage_version: Option<String>,
     ) -> PyResult<Bound<'a, PyAny>> {
         let inner = self_.get_inner()?.clone();
 
@@ -134,8 +126,11 @@ impl Connection {
             builder = builder.storage_options(storage_options);
         }
 
-        if let Some(use_legacy_format) = use_legacy_format {
-            builder = builder.use_legacy_format(use_legacy_format);
+        if let Some(data_storage_version) = data_storage_version.as_ref() {
+            builder = builder.data_storage_version(
+                LanceFileVersion::from_str(data_storage_version)
+                    .map_err(|e| PyValueError::new_err(e.to_string()))?,
+            );
         }
 
         future_into_py(self_.py(), async move {

python/src/index.rs

@@ -98,6 +98,13 @@ pub struct IndexConfig {
     pub columns: Vec<String>,
 }
 
+#[pymethods]
+impl IndexConfig {
+    pub fn __repr__(&self) -> String {
+        format!("Index({}, columns={:?})", self.index_type, self.columns)
+    }
+}
+
 impl From<lancedb::index::IndexConfig> for IndexConfig {
     fn from(value: lancedb::index::IndexConfig) -> Self {
         let index_type = format!("{:?}", value.index_type);

python/src/table.rs

@@ -60,6 +60,16 @@ pub struct Table {
     inner: Option<LanceDbTable>,
 }
 
+#[pymethods]
+impl OptimizeStats {
+    pub fn __repr__(&self) -> String {
+        format!(
+            "OptimizeStats(compaction={:?}, prune={:?})",
+            self.compaction, self.prune
+        )
+    }
+}
+
 impl Table {
     pub(crate) fn new(inner: LanceDbTable) -> Self {
         Self {
@@ -266,6 +276,7 @@ impl Table {
                 .optimize(OptimizeAction::Prune {
                     older_than,
                     delete_unverified: None,
+                    error_if_tagged_old_versions: None,
                 })
                 .await
                 .infer_error()?

rust/ffi/node/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-node"
-version = "0.7.2"
+version = "0.9.0"
 description = "Serverless, low-latency vector database for AI applications"
 license.workspace = true
 edition.workspace = true

rust/ffi/node/src/table.rs

@@ -320,12 +320,19 @@ impl JsTable {
                 .map(|val| val.value(&mut cx))
                 .unwrap_or_default(),
         );
+        let error_if_tagged_old_versions: Option<bool> = Some(
+            cx.argument_opt(2)
+                .and_then(|val| val.downcast::<JsBoolean, _>(&mut cx).ok())
+                .map(|val| val.value(&mut cx))
+                .unwrap_or_default(),
+        );
 
         rt.spawn(async move {
             let stats = table
                 .optimize(OptimizeAction::Prune {
                     older_than: Some(older_than),
                     delete_unverified,
+                    error_if_tagged_old_versions,
                 })
                 .await;
 

rust/lancedb/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.7.2"
+version = "0.9.0"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true
@@ -29,6 +29,7 @@ lance-datafusion.workspace = true
 lance-index = { workspace = true }
 lance-linalg = { workspace = true }
 lance-testing = { workspace = true }
+lance-encoding = { workspace = true }
 pin-project = { workspace = true }
 tokio = { version = "1.23", features = ["rt-multi-thread"] }
 log.workspace = true
@@ -46,11 +47,11 @@ serde_with = { version = "3.8.1" }
 reqwest = { version = "0.11.24", features = ["gzip", "json"], optional = true }
 polars-arrow = { version = ">=0.37,<0.40.0", optional = true }
 polars = { version = ">=0.37,<0.40.0", optional = true }
-hf-hub = {version = "0.3.2", optional = true}
+hf-hub = { version = "0.3.2", optional = true }
 candle-core = { version = "0.6.0", optional = true }
 candle-transformers = { version = "0.6.0", optional = true }
 candle-nn = { version = "0.6.0", optional = true }
-tokenizers = { version = "0.19.1", optional = true}
+tokenizers = { version = "0.19.1", optional = true }
 
 [dev-dependencies]
 tempfile = "3.5.0"
@@ -70,7 +71,13 @@ fp16kernels = ["lance-linalg/fp16kernels"]
 s3-test = []
 openai = ["dep:async-openai", "dep:reqwest"]
 polars = ["dep:polars-arrow", "dep:polars"]
-sentence-transformers = ["dep:hf-hub", "dep:candle-core", "dep:candle-transformers", "dep:candle-nn", "dep:tokenizers"]
+sentence-transformers = [
+    "dep:hf-hub",
+    "dep:candle-core",
+    "dep:candle-transformers",
+    "dep:candle-nn",
+    "dep:tokenizers"
+]
 
 [[example]]
 name = "openai"

rust/lancedb/src/connection.rs

@@ -22,7 +22,7 @@ use std::sync::Arc;
 use arrow_array::{RecordBatchIterator, RecordBatchReader};
 use arrow_schema::SchemaRef;
 use lance::dataset::{ReadParams, WriteMode};
-use lance::io::{ObjectStore, ObjectStoreParams, WrappingObjectStore};
+use lance::io::{ObjectStore, ObjectStoreParams, ObjectStoreRegistry, WrappingObjectStore};
 use object_store::{aws::AwsCredential, local::LocalFileSystem};
 use snafu::prelude::*;
 
@@ -35,6 +35,7 @@ use crate::io::object_store::MirroringObjectStoreWrapper;
 use crate::table::{NativeTable, TableDefinition, WriteOptions};
 use crate::utils::validate_table_name;
 use crate::Table;
+pub use lance_encoding::version::LanceFileVersion;
 
 #[cfg(feature = "remote")]
 use log::warn;
@@ -140,7 +141,7 @@ pub struct CreateTableBuilder<const HAS_DATA: bool, T: IntoArrow> {
     pub(crate) write_options: WriteOptions,
     pub(crate) table_definition: Option<TableDefinition>,
     pub(crate) embeddings: Vec<(EmbeddingDefinition, Arc<dyn EmbeddingFunction>)>,
-    pub(crate) use_legacy_format: bool,
+    pub(crate) data_storage_version: Option<LanceFileVersion>,
 }
 
 // Builder methods that only apply when we have initial data
@@ -154,7 +155,7 @@ impl<T: IntoArrow> CreateTableBuilder<true, T> {
             write_options: WriteOptions::default(),
             table_definition: None,
             embeddings: Vec::new(),
-            use_legacy_format: true,
+            data_storage_version: None,
         }
     }
 
@@ -186,7 +187,7 @@ impl<T: IntoArrow> CreateTableBuilder<true, T> {
             mode: self.mode,
             write_options: self.write_options,
             embeddings: self.embeddings,
-            use_legacy_format: self.use_legacy_format,
+            data_storage_version: self.data_storage_version,
         };
         Ok((data, builder))
     }
@@ -220,7 +221,7 @@ impl CreateTableBuilder<false, NoData> {
             mode: CreateTableMode::default(),
             write_options: WriteOptions::default(),
             embeddings: Vec::new(),
-            use_legacy_format: true,
+            data_storage_version: None,
         }
     }
 
@@ -283,6 +284,14 @@ impl<const HAS_DATA: bool, T: IntoArrow> CreateTableBuilder<HAS_DATA, T> {
         self
     }
 
+    /// Set the data storage version.
+    ///
+    /// The default is `LanceFileVersion::Legacy`.
+    pub fn data_storage_version(mut self, data_storage_version: LanceFileVersion) -> Self {
+        self.data_storage_version = Some(data_storage_version);
+        self
+    }
+
     /// Set to true to use the v1 format for data files
     ///
     /// This is currently defaulted to true and can be set to false to opt-in
@@ -292,8 +301,13 @@ impl<const HAS_DATA: bool, T: IntoArrow> CreateTableBuilder<HAS_DATA, T> {
     ///
     /// Once the new format is stable, the default will change to `false` for
     /// several releases and then eventually this option will be removed.
+    #[deprecated(since = "0.9.0", note = "use data_storage_version instead")]
     pub fn use_legacy_format(mut self, use_legacy_format: bool) -> Self {
-        self.use_legacy_format = use_legacy_format;
+        self.data_storage_version = if use_legacy_format {
+            Some(LanceFileVersion::Legacy)
+        } else {
+            Some(LanceFileVersion::Stable)
+        };
         self
     }
 }
@@ -789,13 +803,14 @@ impl Database {
 
                 let plain_uri = url.to_string();
 
+                let registry = Arc::new(ObjectStoreRegistry::default());
                 let storage_options = options.storage_options.clone();
                 let os_params = ObjectStoreParams {
                     storage_options: Some(storage_options.clone()),
                     ..Default::default()
                 };
                 let (object_store, base_path) =
-                    ObjectStore::from_uri_and_params(&plain_uri, &os_params).await?;
+                    ObjectStore::from_uri_and_params(registry, &plain_uri, &os_params).await?;
                 if object_store.is_local() {
                     Self::try_create_dir(&plain_uri).context(CreateDirSnafu { path: plain_uri })?;
                 }
@@ -961,7 +976,7 @@ impl ConnectionInternal for Database {
         if matches!(&options.mode, CreateTableMode::Overwrite) {
             write_params.mode = WriteMode::Overwrite;
         }
-        write_params.use_legacy_format = options.use_legacy_format;
+        write_params.data_storage_version = options.data_storage_version;
 
         match NativeTable::create(
             &table_uri,

rust/lancedb/src/table.rs

@@ -191,6 +191,8 @@ pub enum OptimizeAction {
         /// Because they may be part of an in-progress transaction, files newer than 7 days old are not deleted by default.
         /// If you are sure that there are no in-progress transactions, then you can set this to True to delete all files older than `older_than`.
         delete_unverified: Option<bool>,
+        /// If true, an error will be returned if there are any old versions that are still tagged.
+        error_if_tagged_old_versions: Option<bool>,
     },
     /// Optimize the indices
     ///
@@ -1079,8 +1081,8 @@ impl NativeTable {
         params: Option<WriteParams>,
         read_consistency_interval: Option<std::time::Duration>,
     ) -> Result<Self> {
+        // Default params uses format v1.
         let params = params.unwrap_or(WriteParams {
-            use_legacy_format: true,
             ..Default::default()
         });
         // patch the params if we have a write store wrapper
@@ -1173,12 +1175,13 @@ impl NativeTable {
         &self,
         older_than: Duration,
         delete_unverified: Option<bool>,
+        error_if_tagged_old_versions: Option<bool>,
     ) -> Result<RemovalStats> {
         Ok(self
             .dataset
             .get_mut()
             .await?
-            .cleanup_old_versions(older_than, delete_unverified)
+            .cleanup_old_versions(older_than, delete_unverified, error_if_tagged_old_versions)
             .await?)
     }
 
@@ -1506,8 +1509,8 @@ impl NativeTable {
         }
 
         let mut dataset = self.dataset.get_mut().await?;
-        let lance_idx_params = lance::index::scalar::ScalarIndexParams {
-            force_index_type: Some(lance::index::scalar::ScalarIndexType::BTree),
+        let lance_idx_params = lance_index::scalar::ScalarIndexParams {
+            force_index_type: Some(lance_index::scalar::ScalarIndexType::BTree),
         };
         dataset
             .create_index(
@@ -1607,6 +1610,9 @@ impl TableInternal for NativeTable {
         let data =
             MaybeEmbedded::try_new(data, self.table_definition().await?, add.embedding_registry)?;
 
+        // Still use the legacy lance format (v1) by default.
+        // We don't want to accidentally switch to v2 format during an add operation.
+        // If the table is already v2 this won't have any effect.
         let mut lance_params = add.write_options.lance_write_params.unwrap_or(WriteParams {
             mode: match add.mode {
                 AddDataMode::Append => WriteMode::Append,
@@ -1628,16 +1634,11 @@ impl TableInternal for NativeTable {
         }
 
         // patch the params if we have a write store wrapper
-        let mut lance_params = match self.store_wrapper.clone() {
+        let lance_params = match self.store_wrapper.clone() {
             Some(wrapper) => lance_params.patch_with_store_wrapper(wrapper)?,
             None => lance_params,
         };
 
-        // Only use the new format if the user passes use_legacy_format=False in while creating
-        // a table with data.  We don't want to accidentally switch to v2 format during an add
-        // operation.  If the table is already v2 this won't have any effect.
-        lance_params.use_legacy_format = true;
-
         self.dataset.ensure_mutable().await?;
         let dataset = Dataset::write(data, &self.uri, Some(lance_params)).await?;
 
@@ -1878,6 +1879,7 @@ impl TableInternal for NativeTable {
                     .optimize(OptimizeAction::Prune {
                         older_than: None,
                         delete_unverified: None,
+                        error_if_tagged_old_versions: None,
                     })
                     .await?
                     .prune;
@@ -1893,11 +1895,13 @@ impl TableInternal for NativeTable {
             OptimizeAction::Prune {
                 older_than,
                 delete_unverified,
+                error_if_tagged_old_versions,
             } => {
                 stats.prune = Some(
                     self.cleanup_old_versions(
                         older_than.unwrap_or(Duration::try_days(7).expect("valid delta")),
                         delete_unverified,
+                        error_if_tagged_old_versions,
                     )
                     .await?,
                 );