reformatted

there's build failure for the rust artifact but the macos arm64 build
for npm publish still passed. So we had a silent failure for 2 releases.
By setting error to immediate this should cause fail immediately.

.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.3.3
+current_version = 0.3.9
 commit = True
 message = Bump version: {current_version} → {new_version}
 tag = True

.github/workflows/node.yml

@@ -11,6 +11,10 @@ on:
       - .github/workflows/node.yml
       - docker-compose.yml
 
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
 env:
   # Disable full debug symbol generation to speed up CI build and keep memory down
   # "1" means line tables only, which is useful for panic tracebacks.

.github/workflows/npm-publish.yml

@@ -38,7 +38,7 @@ jobs:
             node/vectordb-*.tgz
 
   node-macos:
-    runs-on: macos-12
+    runs-on: macos-13
     # Only runs on tags that matches the make-release action
     if: startsWith(github.ref, 'refs/tags/v')
     strategy:

.github/workflows/python.yml

@@ -8,6 +8,11 @@ on:
     paths:
       - python/**
       - .github/workflows/python.yml
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
 jobs:
   linux:
     timeout-minutes: 30
@@ -32,18 +37,19 @@ jobs:
       run: |
         pip install -e .[tests]
         pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
-        pip install pytest pytest-mock black isort
+        pip install pytest pytest-mock ruff
-    - name: Black
+    - name: Lint
-      run: black --check --diff --no-color --quiet .
+      run: ruff format --check .
-    - name: isort
-      run: isort --check --diff --quiet .
     - name: Run tests
       run: pytest -m "not slow" -x -v --durations=30 tests
     - name: doctest
       run: pytest --doctest-modules lancedb
   mac:
     timeout-minutes: 30
-    runs-on: "macos-12"
+    strategy:
+      matrix:
+        mac-runner: [ "macos-13", "macos-13-xlarge" ]
+    runs-on: "${{ matrix.mac-runner }}"
     defaults:
       run:
         shell: bash
@@ -62,8 +68,6 @@ jobs:
         pip install -e .[tests]
         pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
         pip install pytest pytest-mock black
-    - name: Black
-      run: black --check --diff --no-color --quiet .
     - name: Run tests
       run: pytest -m "not slow" -x -v --durations=30 tests
   pydantic1x:
@@ -95,4 +99,4 @@ jobs:
     - name: Run tests
       run: pytest -m "not slow" -x -v --durations=30 tests
     - name: doctest
-      run: pytest --doctest-modules lancedb
+      run: pytest --doctest-modules lancedb

.github/workflows/rust.yml

@@ -10,6 +10,10 @@ on:
       - rust/**
       - .github/workflows/rust.yml
 
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
 env:
   # This env var is used by Swatinem/rust-cache@v2 for the cache
   # key, so we set it to make sure it is always consistent.
@@ -44,8 +48,11 @@ jobs:
     - name: Run tests
       run: cargo test --all-features
   macos:
-    runs-on: macos-12
     timeout-minutes: 30
+    strategy:
+      matrix:
+        mac-runner: [ "macos-13", "macos-13-xlarge" ]
+    runs-on: "${{ matrix.mac-runner }}"
     defaults:
       run:
         shell: bash

Cargo.toml

@@ -5,9 +5,10 @@ exclude = ["python"]
 resolver = "2"
 
 [workspace.dependencies]
-lance = { "version" = "=0.8.7", "features" = ["dynamodb"] }
+lance = { "version" = "=0.8.17", "features" = ["dynamodb"] }
-lance-linalg = { "version" = "=0.8.7" }
+lance-index = { "version" = "=0.8.17" }
-lance-testing = { "version" = "=0.8.7" }
+lance-linalg = { "version" = "=0.8.17" }
+lance-testing = { "version" = "=0.8.17" }
 # Note that this one does not include pyarrow
 arrow = { version = "47.0.0", optional = false }
 arrow-array = "47.0"
@@ -19,7 +20,7 @@ arrow-arith = "47.0"
 arrow-cast = "47.0"
 chrono = "0.4.23"
 half = { "version" = "=2.3.1", default-features = false, features = [
-    "num-traits"
+    "num-traits",
 ] }
 log = "0.4"
 object_store = "0.7.1"

ci/build_macos_artifacts.sh

@@ -1,6 +1,7 @@
 # Builds the macOS artifacts (node binaries).
 # Usage: ./ci/build_macos_artifacts.sh [target]
 # Targets supported: x86_64-apple-darwin aarch64-apple-darwin
+set -e
 
 prebuild_rust() {
     # Building here for the sake of easier debugging.

docs/mkdocs.yml

@@ -1,4 +1,5 @@
 site_name: LanceDB Docs
+site_url: https://lancedb.github.io/lancedb/
 repo_url: https://github.com/lancedb/lancedb
 edit_uri: https://github.com/lancedb/lancedb/tree/main/docs/src
 repo_name: lancedb/lancedb
@@ -79,6 +80,7 @@ nav:
     - Ingest Embedding Functions: embeddings/embedding_functions.md
     - Available Functions: embeddings/default_embedding_functions.md
     - Create Custom Embedding Functions: embeddings/api.md
+    - Example - Calculate CLIP Embeddings with Roboflow Inference: examples/image_embeddings_roboflow.md
     - Example - Multi-lingual semantic search: notebooks/multi_lingual_example.ipynb
     - Example - MultiModal CLIP Embeddings: notebooks/DisappearingEmbeddingFunction.ipynb
   - 🔍 Python full-text search: fts.md
@@ -144,14 +146,13 @@ nav:
   - Serverless Chatbot from any website: examples/serverless_website_chatbot.md
   - TransformersJS Embedding Search: examples/transformerjs_embedding_search_nodejs.md
 - API references:
-  - Python API: python/python.md
+  - OSS Python API: python/python.md
+  - SaaS Python API: python/saas-python.md
   - Javascript API: javascript/modules.md
 - LanceDB Cloud↗: https://noteforms.com/forms/lancedb-mailing-list-cloud-kty1o5?notionforms=1&utm_source=notionforms
 
 extra_css:
   - styles/global.css
-extra_javascript:
-  - scripts/posthog.js
 
 extra:
   analytics:

docs/src/ann_indexes.md

@@ -71,9 +71,41 @@ a single PQ code.
 ### Use GPU to build vector index
 
 Lance Python SDK has experimental GPU support for creating IVF index.
+Using GPU for index creation requires [PyTorch>2.0](https://pytorch.org/) being installed.
+
 You can specify the GPU device to train IVF partitions via
 
-- **accelerator**: Specify to `"cuda"`` to enable GPU training.
+- **accelerator**: Specify to ``cuda`` or ``mps`` (on Apple Silicon) to enable GPU training.
+
+=== "Linux"
+
+     <!-- skip-test -->
+     ``` { .python .copy }
+     # Create index using CUDA on Nvidia GPUs.
+     tbl.create_index(
+          num_partitions=256,
+          num_sub_vectors=96,
+          accelerator="cuda"
+     )
+     ```
+
+=== "Macos"
+
+     <!-- skip-test -->
+     ```python
+     # Create index using MPS on Apple Silicon.
+     tbl.create_index(
+          num_partitions=256,
+          num_sub_vectors=96,
+          accelerator="mps"
+     )
+     ```
+
+Trouble shootings:
+
+If you see ``AssertionError: Torch not compiled with CUDA enabled``, you need to [install
+PyTorch with CUDA support](https://pytorch.org/get-started/locally/).
+
 
 ## Querying an ANN Index
 

docs/src/embeddings/default_embedding_functions.md

@@ -1,7 +1,9 @@
 There are various Embedding functions available out of the box with lancedb. We're working on supporting other popular embedding APIs.
 
 ## Text Embedding Functions
-Here are the text embedding functions registered by default
+Here are the text embedding functions registered by default.
+Embedding functions have inbuilt rate limit handler wrapper for source and query embedding function calls that retry with exponential standoff. 
+Each `EmbeddingFunction` implementation automatically takes `max_retries` as an argument which has the deafult value of 7.
 
 ### Sentence Transformers
 Here are the parameters that you can set when registering a `sentence-transformers` object, and their default values:
@@ -66,6 +68,56 @@ actual = table.search(query).limit(1).to_pydantic(Words)[0]
 print(actual.text)
 ```
 
+### Instructor Embeddings
+Instructor is an instruction-finetuned text embedding model that can generate text embeddings tailored to any task (e.g., classification, retrieval, clustering, text evaluation, etc.) and domains (e.g., science, finance, etc.) by simply providing the task instruction, without any finetuning
+
+If you want to calculate customized embeddings for specific sentences, you may follow the unified template to write instructions:
+
+                          Represent the `domain` `text_type` for `task_objective`:
+
+* `domain` is optional, and it specifies the domain of the text, e.g., science, finance, medicine, etc.
+* `text_type` is required, and it specifies the encoding unit, e.g., sentence, document, paragraph, etc.
+* `task_objective` is optional, and it specifies the objective of embedding, e.g., retrieve a document, classify the sentence, etc.
+
+More information about the model can be found here - https://github.com/xlang-ai/instructor-embedding
+
+| Argument | Type | Default | Description |
+|---|---|---|---|
+| `name` | `str` | "hkunlp/instructor-base" | The name of the model to use |
+| `batch_size` | `int` | `32` | The batch size to use when generating embeddings |
+| `device` | `str` | `"cpu"` | The device to use when generating embeddings |
+| `show_progress_bar` | `bool` | `True` | Whether to show a progress bar when generating embeddings |
+| `normalize_embeddings` | `bool` | `True` | Whether to normalize the embeddings |
+| `quantize` | `bool` | `False` | Whether to quantize the model |
+| `source_instruction` | `str` | `"represent the docuement for retreival"` | The instruction for the source column |
+| `query_instruction` | `str` | `"represent the document for retreiving the most similar documents"` | The instruction for the query |
+
+
+
+```python
+import lancedb
+from lancedb.pydantic import LanceModel, Vector
+from lancedb.embeddings import get_registry, InstuctorEmbeddingFunction
+
+instructor = get_registry().get("instructor").create(
+                            source_instruction="represent the docuement for retreival",
+                            query_instruction="represent the document for retreiving the most similar documents"
+                            )
+
+class Schema(LanceModel):
+    vector: Vector(instructor.ndims()) = instructor.VectorField()
+    text: str = instructor.SourceField()
+
+db = lancedb.connect("~/.lancedb")
+tbl = db.create_table("test", schema=Schema, mode="overwrite")
+
+texts = [{"text": "Capitalism has been dominant in the Western world since the end of feudalism, but most feel[who?] that..."},
+        {"text": "The disparate impact theory is especially controversial under the Fair Housing Act because the Act..."},
+        {"text": "Disparate impact in United States labor law refers to practices in employment, housing, and other areas that.."}]
+
+tbl.add(texts)
+```
+
 ## Multi-modal embedding functions
 Multi-modal embedding functions allow you query your table using both images and text.
 

docs/src/embeddings/embedding_functions.md

@@ -57,6 +57,19 @@ query_image = Image.open(p)
 table.search(query_image)
 
 ```
+### Rate limit Handling
+`EmbeddingFunction` class wraps the calls for source and query embedding generation inside a rate limit handler that retries the requests with exponential backoff after successive failures. By default the maximum retires is set to 7. You can tune it by setting it to a different number or disable it by setting it to 0.
+Example
+----
+
+```python
+clip = registry.get("open-clip").create() # Defaults to 7 max retries
+clip = registry.get("open-clip").create(max_retries=10) # Increase max retries to 10
+clip = registry.get("open-clip").create(max_retries=0) # Retries disabled
+````
+
+NOTE:
+Embedding functions can also fail due to other errors that have nothing to do with rate limits. This is why the error is also logged.
 
 ### A little fun with PyDantic
 LanceDB is integrated with PyDantic. Infact we've used the integration in the above example to define the schema. It is also being used behing the scene by the embdding function API to ingest useful information as table metadata.

docs/src/examples/image_embeddings_roboflow.md

@@ -0,0 +1,165 @@
+# How to Load Image Embeddings into LanceDB
+
+With the rise of Large Multimodal Models (LMMs) such as [GPT-4 Vision](https://blog.roboflow.com/gpt-4-vision/), the need for storing image embeddings is growing. The most effective way to store text and image embeddings is in a vector database such as LanceDB. Vector databases are a special kind of data store that enables efficient search over stored embeddings. 
+
+[CLIP](https://blog.roboflow.com/openai-clip/), a multimodal model developed by OpenAI, is commonly used to calculate image embeddings. These embeddings can then be used with a vector database to build a semantic search engine that you can query using images or text. For example, you could use LanceDB and CLIP embeddings to build a search engine for a database of folders.
+
+In this guide, we are going to show you how to use Roboflow Inference to load image embeddings into LanceDB. Without further ado, let’s get started!
+
+## Step #1: Install Roboflow Inference
+
+[Roboflow Inference](https://inference.roboflow.com) enables you to run state-of-the-art computer vision models with minimal configuration. Inference supports a range of models, from fine-tuned object detection, classification, and segmentation models to foundation models like CLIP. We will use Inference to calculate CLIP image embeddings.
+
+Inference provides a HTTP API through which you can run vision models.
+
+Inference powers the Roboflow hosted API, and is available as an open source utility. In this guide, we are going to run Inference locally, which enables you to calculate CLIP embeddings on your own hardware. We will also show you how to use the hosted Roboflow CLIP API, which is ideal if you need to scale and do not want to manage a system for calculating embeddings.
+
+To get started, first install the Inference CLI:
+
+```
+pip install inference-cli
+```
+
+Next, install Docker. Refer to the official Docker installation instructions for your operating system to get Docker set up. Once Docker is ready, you can start Inference using the following command:
+
+```
+inference server start
+```
+
+An Inference server will start running at ‘http://localhost:9001’.
+
+## Step #2: Set Up a LanceDB Vector Database
+
+Now that we have Inference running, we can set up a LanceDB vector database. You can run LanceDB in JavaScript and Python. For this guide, we will use the Python API. But, you can take the HTTP requests we make below and change them to JavaScript if required.
+
+For this guide, we are going to search the [COCO 128 dataset](https://universe.roboflow.com/team-roboflow/coco-128), which contains a wide range of objects. The variability in objects present in this dataset makes it a good dataset to demonstrate the capabilities of vector search. If you want to use this dataset, you can download [COCO 128 from Roboflow Universe](https://universe.roboflow.com/team-roboflow/coco-128). With that said, you can search whatever folder of images you want.
+
+Once you have a dataset ready, install LanceDB with the following command:
+
+```
+pip install lancedb
+```
+
+We also need to install a specific commit of `tantivy`, a dependency of the LanceDB full text search engine we will use later in this guide:
+
+```
+pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
+```
+
+Create a new Python file and add the following code:
+
+```python
+import cv2
+import supervision as sv
+import requests
+
+import lancedb
+
+db = lancedb.connect("./embeddings")
+
+IMAGE_DIR = "images/"
+API_KEY = os.environ.get("ROBOFLOW_API_KEY")
+SERVER_URL = "http://localhost:9001"
+
+results = []
+
+for i, image in enumerate(os.listdir(IMAGE_DIR)):
+    infer_clip_payload = {
+        #Images can be provided as urls or as base64 encoded strings
+        "image": {
+            "type": "base64",
+            "value": base64.b64encode(open(IMAGE_DIR + image, "rb").read()).decode("utf-8"),
+        },
+    }
+
+    res = requests.post(
+        f"{SERVER_URL}/clip/embed_image?api_key={API_KEY}",
+        json=infer_clip_payload,
+    )
+
+    embeddings = res.json()['embeddings']
+
+    print("Calculated embedding for image: ", image)
+
+    image = {"vector": embeddings[0], "name": os.path.join(IMAGE_DIR, image)}
+
+    results.append(image)
+
+tbl = db.create_table("images", data=results)
+
+tbl.create_fts_index("name")
+```
+
+To use the code above, you will need a Roboflow API key. [Learn how to retrieve a Roboflow API key](https://docs.roboflow.com/api-reference/authentication#retrieve-an-api-key). Run the following command to set up your API key in your environment:
+
+```
+export ROBOFLOW_API_KEY=""
+```
+
+Replace the `IMAGE_DIR` value with the folder in which you are storing the images for which you want to calculate embeddings. If you want to use the Roboflow CLIP API to calculate embeddings, replace the `SERVER_URL` value with `https://infer.roboflow.com`.
+
+Run the script above to create a new LanceDB database. This database will be stored on your local machine. The database will be called `embeddings` and the table will be called `images`.
+
+The script above calculates all embeddings for a folder then creates a new table. To add additional images, use the following code:
+
+```python
+def make_batches():
+    for i in range(5):
+        yield [
+                {"vector": [3.1, 4.1], "name": "image1.png"},
+                {"vector": [5.9, 26.5], "name": "image2.png"}
+            ]
+
+tbl = db.open_table("images")
+tbl.add(make_batches())
+```
+
+Replacing the `make_batches()` function with code to load embeddings for images.
+
+## Step #3: Run a Search Query
+
+We are now ready to run a search query. To run a search query, we need a text embedding that represents a text query. We can use this embedding to search our LanceDB database for an entry.
+
+Let’s calculate a text embedding for the query “cat”, then run a search query:
+
+```python
+infer_clip_payload = {
+    "text": "cat",
+}
+
+res = requests.post(
+    f"{SERVER_URL}/clip/embed_text?api_key={API_KEY}",
+    json=infer_clip_payload,
+)
+
+embeddings = res.json()['embeddings']
+
+df = tbl.search(embeddings[0]).limit(3).to_list()
+
+print("Results:")
+
+for i in df:
+    print(i["name"])
+```
+
+This code will search for the three images most closely related to the prompt “cat”. The names of the most similar three images will be printed to the console. Here are the three top results:
+
+```
+dataset/images/train/000000000650_jpg.rf.1b74ba165c5a3513a3211d4a80b69e1c.jpg
+dataset/images/train/000000000138_jpg.rf.af439ef1c55dd8a4e4b142d186b9c957.jpg
+dataset/images/train/000000000165_jpg.rf.eae14d5509bf0c9ceccddbb53a5f0c66.jpg
+```
+
+Let’s open the top image:
+
+![Cat](https://media.roboflow.com/cat_lancedb.jpg)
+
+The top image was a cat. Our search was successful.
+
+## Conclusion
+
+LanceDB is a vector database that you can use to store and efficiently search your image embeddings. You can use Roboflow Inference, a scalable computer vision inference server, to calculate CLIP embeddings that you can store in LanceDB.
+
+You can use Inference and LanceDB together to build a range of applications with image embeddings, from a media search engine to a retrieval-augmented generation pipeline for use with LMMs.
+
+To learn more about Inference and its capabilities, refer to the Inference documentation.

docs/src/python/python.md

@@ -22,8 +22,6 @@ pip install lancedb
 
 ::: lancedb.query.LanceQueryBuilder
 
-::: lancedb.query.LanceFtsQueryBuilder
-
 ## Embeddings
 
 ::: lancedb.embeddings.registry.EmbeddingFunctionRegistry
@@ -56,7 +54,7 @@ pip install lancedb
 
 ## Utilities
 
-::: lancedb.vector
+::: lancedb.schema.vector
 
 ## Integrations
 

docs/src/python/saas-python.md

@@ -0,0 +1,18 @@
+# LanceDB Python API Reference
+
+## Installation
+
+```shell
+pip install lancedb
+```
+
+## Connection
+
+::: lancedb.connect
+
+::: lancedb.remote.db.RemoteDBConnection
+
+## Table
+
+::: lancedb.remote.table.RemoteTable
+

docs/src/robots.txt

@@ -0,0 +1 @@
+User-agent: *

docs/test/md_testing.py

@@ -18,29 +18,45 @@ python_file = ".py"
 python_folder = "python"
 
 files = glob.glob(glob_string, recursive=True)
-excluded_files = [f for excluded_glob in excluded_globs for f in glob.glob(excluded_glob, recursive=True)]
+excluded_files = [
+    f
+    for excluded_glob in excluded_globs
+    for f in glob.glob(excluded_glob, recursive=True)
+]
+
 
 def yield_lines(lines: Iterator[str], prefix: str, suffix: str):
     in_code_block = False
     # Python code has strict indentation
     strip_length = 0
+    skip_test = False
     for line in lines:
+        if "skip-test" in line:
+            skip_test = True
         if line.strip().startswith(prefix + python_prefix):
             in_code_block = True
             strip_length = len(line) - len(line.lstrip())
         elif in_code_block and line.strip().startswith(suffix):
             in_code_block = False
-            yield "\n"
+            if not skip_test:
+                yield "\n"
+            skip_test = False
         elif in_code_block:
-            yield line[strip_length:]
+            if not skip_test:
+                yield line[strip_length:]
 
 for file in filter(lambda file: file not in excluded_files, files):
     with open(file, "r") as f:
         lines = list(yield_lines(iter(f), "```", "```"))
 
     if len(lines) > 0:
-        out_path = Path(python_folder) / Path(file).name.strip(".md") / (Path(file).name.strip(".md") + python_file)
+        print(lines)
+        out_path = (
+            Path(python_folder)
+            / Path(file).name.strip(".md")
+            / (Path(file).name.strip(".md") + python_file)
+        )
         print(out_path)
         out_path.parent.mkdir(exist_ok=True, parents=True)
         with open(out_path, "w") as out:
-            out.writelines(lines)
+            out.writelines(lines)

node/README.md

@@ -10,7 +10,7 @@ npm install vectordb
 
 This will download the appropriate native library for your platform. We currently
 support x86_64 Linux, aarch64 Linux, Intel MacOS, and ARM (M1/M2) MacOS. We do not
-yet support Windows or musl-based Linux (such as Alpine Linux).
+yet support musl-based Linux (such as Alpine Linux).
 
 ## Usage
 

node/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "vectordb",
-  "version": "0.3.3",
+  "version": "0.3.9",
   "lockfileVersion": 2,
   "requires": true,
   "packages": {
     "": {
       "name": "vectordb",
-      "version": "0.3.3",
+      "version": "0.3.9",
       "cpu": [
         "x64",
         "arm64"
@@ -53,11 +53,11 @@
         "uuid": "^9.0.0"
       },
       "optionalDependencies": {
-        "@lancedb/vectordb-darwin-arm64": "0.3.3",
+        "@lancedb/vectordb-darwin-arm64": "0.3.9",
-        "@lancedb/vectordb-darwin-x64": "0.3.3",
+        "@lancedb/vectordb-darwin-x64": "0.3.9",
-        "@lancedb/vectordb-linux-arm64-gnu": "0.3.3",
+        "@lancedb/vectordb-linux-arm64-gnu": "0.3.9",
-        "@lancedb/vectordb-linux-x64-gnu": "0.3.3",
+        "@lancedb/vectordb-linux-x64-gnu": "0.3.9",
-        "@lancedb/vectordb-win32-x64-msvc": "0.3.3"
+        "@lancedb/vectordb-win32-x64-msvc": "0.3.9"
       }
     },
     "node_modules/@apache-arrow/ts": {
@@ -316,22 +316,10 @@
         "@jridgewell/sourcemap-codec": "^1.4.10"
       }
     },
-    "node_modules/@lancedb/vectordb-darwin-arm64": {
-      "version": "0.3.3",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.3.3.tgz",
-      "integrity": "sha512-nvyj7xNX2/wb/PH5TjyhLR/NQ1jVuoBw2B5UaSg7qf8Tnm5SSXWQ7F25RVKcKwh72fz1qB+CWW24ftZnRzbT/Q==",
-      "cpu": [
-        "arm64"
-      ],
-      "optional": true,
-      "os": [
-        "darwin"
-      ]
-    },
     "node_modules/@lancedb/vectordb-darwin-x64": {
-      "version": "0.3.3",
+      "version": "0.3.9",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.3.3.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.3.9.tgz",
-      "integrity": "sha512-7CW+nILyPHp6cua0Rl0xaTDWw/vajEn/jCsEjFYgDmE+rtf5Z5Fum41FxR9C2TtIAvUK+nWb5mkYeOLqU6vRvg==",
+      "integrity": "sha512-4xXQoPheyIl1P5kRoKmZtaAHFrYdL9pw5yq+r6ewIx0TCemN4LSvzSUTqM5nZl3QPU8FeL0CGD8Gt2gMU0HQ2A==",
       "cpu": [
         "x64"
       ],
@@ -341,9 +329,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-linux-arm64-gnu": {
-      "version": "0.3.3",
+      "version": "0.3.9",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.3.3.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.3.9.tgz",
-      "integrity": "sha512-MmhwbacKxZPkLwwOqysVY8mUb8lFoyFIPlYhSLV4xS1C8X4HWALljIul1qMl1RYudp9Uc3PsOzRexl+OvCGfUw==",
+      "integrity": "sha512-WIxCZKnLeSlz0PGURtKSX6hJ4CYE2o5P+IFmmuWOWB1uNapQu6zOpea6rNxcRFHUA0IJdO02lVxVfn2hDX4SMg==",
       "cpu": [
         "arm64"
       ],
@@ -353,9 +341,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-linux-x64-gnu": {
-      "version": "0.3.3",
+      "version": "0.3.9",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.3.3.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.3.9.tgz",
-      "integrity": "sha512-OrNlsKi/QPw59Po040oRKn8IuqFEk4upc/4FaFKqVkcmQjjZrMg5Kgy9ZfWIhHdAnWXXggZZIPArpt0X1B0ceA==",
+      "integrity": "sha512-bQbcV9adKzYbJLNzDjk9OYsMnT2IjmieLfb4IQ1hj5IUoWfbg80Bd0+gZUnrmrhG6fe56TIriFZYQR9i7TSE9Q==",
       "cpu": [
         "x64"
       ],
@@ -365,9 +353,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-win32-x64-msvc": {
-      "version": "0.3.3",
+      "version": "0.3.9",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.3.3.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.3.9.tgz",
-      "integrity": "sha512-lIT0A7a6eqX51IfGyhECtpXXgsr//kgbd+HZbcCdPy2GMmNezSch/7V22zExDSpF32hX8WfgcTLYCVWVilggDQ==",
+      "integrity": "sha512-7EXI7P1QvAfgJNPWWBMDOkoJ696gSBAClcyEJNYg0JV21jVFZRwJVI3bZXflesWduFi/mTuzPkFFA68us1u19A==",
       "cpu": [
         "x64"
       ],
@@ -4868,34 +4856,28 @@
         "@jridgewell/sourcemap-codec": "^1.4.10"
       }
     },
-    "@lancedb/vectordb-darwin-arm64": {
-      "version": "0.3.3",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.3.3.tgz",
-      "integrity": "sha512-nvyj7xNX2/wb/PH5TjyhLR/NQ1jVuoBw2B5UaSg7qf8Tnm5SSXWQ7F25RVKcKwh72fz1qB+CWW24ftZnRzbT/Q==",
-      "optional": true
-    },
     "@lancedb/vectordb-darwin-x64": {
-      "version": "0.3.3",
+      "version": "0.3.9",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.3.3.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.3.9.tgz",
-      "integrity": "sha512-7CW+nILyPHp6cua0Rl0xaTDWw/vajEn/jCsEjFYgDmE+rtf5Z5Fum41FxR9C2TtIAvUK+nWb5mkYeOLqU6vRvg==",
+      "integrity": "sha512-4xXQoPheyIl1P5kRoKmZtaAHFrYdL9pw5yq+r6ewIx0TCemN4LSvzSUTqM5nZl3QPU8FeL0CGD8Gt2gMU0HQ2A==",
       "optional": true
     },
     "@lancedb/vectordb-linux-arm64-gnu": {
-      "version": "0.3.3",
+      "version": "0.3.9",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.3.3.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.3.9.tgz",
-      "integrity": "sha512-MmhwbacKxZPkLwwOqysVY8mUb8lFoyFIPlYhSLV4xS1C8X4HWALljIul1qMl1RYudp9Uc3PsOzRexl+OvCGfUw==",
+      "integrity": "sha512-WIxCZKnLeSlz0PGURtKSX6hJ4CYE2o5P+IFmmuWOWB1uNapQu6zOpea6rNxcRFHUA0IJdO02lVxVfn2hDX4SMg==",
       "optional": true
     },
     "@lancedb/vectordb-linux-x64-gnu": {
-      "version": "0.3.3",
+      "version": "0.3.9",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.3.3.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.3.9.tgz",
-      "integrity": "sha512-OrNlsKi/QPw59Po040oRKn8IuqFEk4upc/4FaFKqVkcmQjjZrMg5Kgy9ZfWIhHdAnWXXggZZIPArpt0X1B0ceA==",
+      "integrity": "sha512-bQbcV9adKzYbJLNzDjk9OYsMnT2IjmieLfb4IQ1hj5IUoWfbg80Bd0+gZUnrmrhG6fe56TIriFZYQR9i7TSE9Q==",
       "optional": true
     },
     "@lancedb/vectordb-win32-x64-msvc": {
-      "version": "0.3.3",
+      "version": "0.3.9",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.3.3.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.3.9.tgz",
-      "integrity": "sha512-lIT0A7a6eqX51IfGyhECtpXXgsr//kgbd+HZbcCdPy2GMmNezSch/7V22zExDSpF32hX8WfgcTLYCVWVilggDQ==",
+      "integrity": "sha512-7EXI7P1QvAfgJNPWWBMDOkoJ696gSBAClcyEJNYg0JV21jVFZRwJVI3bZXflesWduFi/mTuzPkFFA68us1u19A==",
       "optional": true
     },
     "@neon-rs/cli": {

node/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vectordb",
-  "version": "0.3.3",
+  "version": "0.3.9",
   "description": " Serverless, low-latency vector database for AI applications",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
@@ -81,10 +81,10 @@
     }
   },
   "optionalDependencies": {
-    "@lancedb/vectordb-darwin-arm64": "0.3.3",
+    "@lancedb/vectordb-darwin-arm64": "0.3.9",
-    "@lancedb/vectordb-darwin-x64": "0.3.3",
+    "@lancedb/vectordb-darwin-x64": "0.3.9",
-    "@lancedb/vectordb-linux-arm64-gnu": "0.3.3",
+    "@lancedb/vectordb-linux-arm64-gnu": "0.3.9",
-    "@lancedb/vectordb-linux-x64-gnu": "0.3.3",
+    "@lancedb/vectordb-linux-x64-gnu": "0.3.9",
-    "@lancedb/vectordb-win32-x64-msvc": "0.3.3"
+    "@lancedb/vectordb-win32-x64-msvc": "0.3.9"
   }
 }

node/src/index.ts

@@ -23,7 +23,7 @@ import { Query } from './query'
 import { isEmbeddingFunction } from './embedding/embedding_function'
 
 // eslint-disable-next-line @typescript-eslint/no-var-requires
-const { databaseNew, databaseTableNames, databaseOpenTable, databaseDropTable, tableCreate, tableAdd, tableCreateVectorIndex, tableCountRows, tableDelete, tableCleanupOldVersions, tableCompactFiles } = require('../native.js')
+const { databaseNew, databaseTableNames, databaseOpenTable, databaseDropTable, tableCreate, tableAdd, tableCreateVectorIndex, tableCountRows, tableDelete, tableCleanupOldVersions, tableCompactFiles, tableListIndices, tableIndexStats } = require('../native.js')
 
 export { Query }
 export type { EmbeddingFunction }
@@ -260,6 +260,27 @@ export interface Table<T = number[]> {
    * ```
    */
   delete: (filter: string) => Promise<void>
+
+  /**
+   * List the indicies on this table.
+   */
+  listIndices: () => Promise<VectorIndex[]>
+
+  /**
+   * Get statistics about an index.
+   */
+  indexStats: (indexUuid: string) => Promise<IndexStats>
+}
+
+export interface VectorIndex {
+  columns: string[]
+  name: string
+  uuid: string
+}
+
+export interface IndexStats {
+  numIndexedRows: number | null
+  numUnindexedRows: number | null
 }
 
 /**
@@ -502,6 +523,14 @@ export class LocalTable<T = number[]> implements Table<T> {
         return res.metrics
       })
   }
+
+  async listIndices (): Promise<VectorIndex[]> {
+    return tableListIndices.call(this._tbl)
+  }
+
+  async indexStats (indexUuid: string): Promise<IndexStats> {
+    return tableIndexStats.call(this._tbl, indexUuid)
+  }
 }
 
 export interface CleanupStats {

node/src/query.ts

@@ -32,6 +32,7 @@ export class Query<T = number[]> {
   private _select?: string[]
   private _filter?: string
   private _metricType?: MetricType
+  private _prefilter: boolean
   protected readonly _embeddings?: EmbeddingFunction<T>
 
   constructor (query: T, tbl?: any, embeddings?: EmbeddingFunction<T>) {
@@ -44,6 +45,7 @@ export class Query<T = number[]> {
     this._filter = undefined
     this._metricType = undefined
     this._embeddings = embeddings
+    this._prefilter = false
   }
 
   /***
@@ -102,6 +104,11 @@ export class Query<T = number[]> {
     return this
   }
 
+  prefilter (value: boolean): Query<T> {
+    this._prefilter = value
+    return this
+  }
+
   /**
      * Execute the query and return the results as an Array of Objects
      */

node/src/remote/client.ts

@@ -38,6 +38,7 @@ export class HttpLancedbClient {
     vector: number[],
     k: number,
     nprobes: number,
+    prefilter: boolean,
     refineFactor?: number,
     columns?: string[],
     filter?: string
@@ -50,7 +51,8 @@ export class HttpLancedbClient {
                 nprobes,
                 refineFactor,
                 columns,
-                filter
+                filter,
+                prefilter
               },
               {
                 headers: {
@@ -63,6 +65,9 @@ export class HttpLancedbClient {
               }
     ).catch((err) => {
       console.error('error: ', err)
+      if (err.response === undefined) {
+        throw new Error(`Network Error: ${err.message as string}`)
+      }
       return err.response
     })
     if (response.status !== 200) {
@@ -86,13 +91,17 @@ export class HttpLancedbClient {
       {
         headers: {
           'Content-Type': 'application/json',
-          'x-api-key': this._apiKey()
+          'x-api-key': this._apiKey(),
+          ...(this._dbName !== undefined ? { 'x-lancedb-database': this._dbName } : {})
         },
         params,
         timeout: 10000
       }
     ).catch((err) => {
       console.error('error: ', err)
+      if (err.response === undefined) {
+        throw new Error(`Network Error: ${err.message as string}`)
+      }
       return err.response
     })
     if (response.status !== 200) {
@@ -128,6 +137,9 @@ export class HttpLancedbClient {
         }
     ).catch((err) => {
       console.error('error: ', err)
+      if (err.response === undefined) {
+        throw new Error(`Network Error: ${err.message as string}`)
+      }
       return err.response
     })
     if (response.status !== 200) {

node/src/remote/index.ts

@@ -14,7 +14,9 @@
 
 import {
   type EmbeddingFunction, type Table, type VectorIndexParams, type Connection,
-  type ConnectionOptions, type CreateTableOptions, type WriteOptions
+  type ConnectionOptions, type CreateTableOptions, type VectorIndex,
+  type WriteOptions,
+  type IndexStats
 } from '../index'
 import { Query } from '../query'
 
@@ -152,6 +154,7 @@ export class RemoteQuery<T = number[]> extends Query<T> {
       queryVector,
       (this as any)._limit,
       (this as any)._nprobes,
+      (this as any)._prefilter,
       (this as any)._refineFactor,
       (this as any)._select,
       (this as any)._filter
@@ -235,10 +238,28 @@ export class RemoteTable<T = number[]> implements Table<T> {
   }
 
   async countRows (): Promise<number> {
-    throw new Error('Not implemented')
+    const result = await this._client.post(`/v1/table/${this._name}/describe/`)
+    return result.data?.stats?.num_rows
   }
 
   async delete (filter: string): Promise<void> {
     await this._client.post(`/v1/table/${this._name}/delete/`, { predicate: filter })
   }
+
+  async listIndices (): Promise<VectorIndex[]> {
+    const results = await this._client.post(`/v1/table/${this._name}/index/list/`)
+    return results.data.indexes?.map((index: any) => ({
+      columns: index.columns,
+      name: index.index_name,
+      uuid: index.index_uuid
+    }))
+  }
+
+  async indexStats (indexUuid: string): Promise<IndexStats> {
+    const results = await this._client.post(`/v1/table/${this._name}/index/${indexUuid}/stats/`)
+    return {
+      numIndexedRows: results.data.num_indexed_rows,
+      numUnindexedRows: results.data.num_unindexed_rows
+    }
+  }
 }

node/src/test/test.ts

@@ -102,6 +102,20 @@ describe('LanceDB client', function () {
       assertResults(results)
     })
 
+    it('should correctly process prefilter/postfilter', async function () {
+      const uri = await createTestDB(16, 300)
+      const con = await lancedb.connect(uri)
+      const table = await con.openTable('vectors')
+      await table.createIndex({ type: 'ivf_pq', column: 'vector', num_partitions: 2, max_iters: 2, num_sub_vectors: 2 })
+      // post filter should return less than the limit
+      let results = await table.search(new Array(16).fill(0.1)).limit(10).filter('id >= 10').prefilter(false).execute()
+      assert.isTrue(results.length < 10)
+
+      // pre filter should return exactly the limit
+      results = await table.search(new Array(16).fill(0.1)).limit(10).filter('id >= 10').prefilter(true).execute()
+      assert.isTrue(results.length === 10)
+    })
+
     it('select only a subset of columns', async function () {
       const uri = await createTestDB()
       const con = await lancedb.connect(uri)
@@ -282,7 +296,8 @@ describe('LanceDB client', function () {
       )
       const table = await con.createTable({ name: 'vectors', schema })
       await table.add([{ vector: Array(128).fill(0.1) }])
-      await table.delete('vector IS NOT NULL')
+      // https://github.com/lancedb/lance/issues/1635
+      await table.delete('true')
       const result = await table.search(Array(128).fill(0.1)).execute()
       assert.isEmpty(result)
     })
@@ -328,6 +343,24 @@ describe('LanceDB client', function () {
       const createIndex = table.createIndex({ type: 'ivf_pq', column: 'name', num_partitions: -1, max_iters: 2, num_sub_vectors: 2 })
       await expect(createIndex).to.be.rejectedWith('num_partitions: must be > 0')
     })
+
+    it('should be able to list index and stats', async function () {
+      const uri = await createTestDB(32, 300)
+      const con = await lancedb.connect(uri)
+      const table = await con.openTable('vectors')
+      await table.createIndex({ type: 'ivf_pq', column: 'vector', num_partitions: 2, max_iters: 2, num_sub_vectors: 2 })
+
+      const indices = await table.listIndices()
+      expect(indices).to.have.lengthOf(1)
+      expect(indices[0].name).to.equal('vector_idx')
+      expect(indices[0].uuid).to.not.be.equal(undefined)
+      expect(indices[0].columns).to.have.lengthOf(1)
+      expect(indices[0].columns[0]).to.equal('vector')
+
+      const stats = await table.indexStats(indices[0].uuid)
+      expect(stats.numIndexedRows).to.equal(300)
+      expect(stats.numUnindexedRows).to.equal(0)
+    }).timeout(50_000)
   })
 
   describe('when using a custom embedding function', function () {
@@ -378,6 +411,40 @@ describe('LanceDB client', function () {
   })
 })
 
+describe('Remote LanceDB client', function () {
+  describe('when the server is not reachable', function () {
+    it('produces a network error', async function () {
+      const con = await lancedb.connect({
+        uri: 'db://test-1234',
+        region: 'asdfasfasfdf',
+        apiKey: 'some-api-key'
+      })
+
+      // GET
+      try {
+        await con.tableNames()
+      } catch (err) {
+        expect(err).to.have.property('message', 'Network Error: getaddrinfo ENOTFOUND test-1234.asdfasfasfdf.api.lancedb.com')
+      }
+
+      // POST
+      try {
+        await con.createTable({ name: 'vectors', schema: new Schema([]) })
+      } catch (err) {
+        expect(err).to.have.property('message', 'Network Error: getaddrinfo ENOTFOUND test-1234.asdfasfasfdf.api.lancedb.com')
+      }
+
+      // Search
+      const table = await con.openTable('vectors')
+      try {
+        await table.search([0.1, 0.3]).execute()
+      } catch (err) {
+        expect(err).to.have.property('message', 'Network Error: getaddrinfo ENOTFOUND test-1234.asdfasfasfdf.api.lancedb.com')
+      }
+    })
+  })
+})
+
 describe('Query object', function () {
   it('sets custom parameters', async function () {
     const query = new Query([0.1, 0.3])

python/.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.3.2
+current_version = 0.3.4
 commit = True
 message = [python] Bump version: {current_version} → {new_version}
 tag = True

python/lancedb/__init__.py

@@ -16,17 +16,18 @@ from typing import Optional
 
 __version__ = importlib.metadata.version("lancedb")
 
-from .db import URI, DBConnection, LanceDBConnection
+from .common import URI
+from .db import DBConnection, LanceDBConnection
 from .remote.db import RemoteDBConnection
-from .schema import vector
+from .schema import vector  # noqa: F401
-from .utils import sentry_log
+from .utils import sentry_log  # noqa: F401
 
 
 def connect(
     uri: URI,
     *,
     api_key: Optional[str] = None,
-    region: str = "us-west-2",
+    region: str = "us-east-1",
     host_override: Optional[str] = None,
 ) -> DBConnection:
     """Connect to a LanceDB database.
@@ -38,7 +39,7 @@ def connect(
     api_key: str, optional
         If presented, connect to LanceDB cloud.
         Otherwise, connect to a database on file system or cloud storage.
-    region: str, default "us-west-2"
+    region: str, default "us-east-1"
         The region to use for LanceDB Cloud.
     host_override: str, optional
         The override url for LanceDB Cloud.

python/lancedb/conftest.py

@@ -1,4 +1,6 @@
 import os
+import time
+from typing import Any
 
 import numpy as np
 import pytest
@@ -38,3 +40,26 @@ class MockTextEmbeddingFunction(TextEmbeddingFunction):
 
     def ndims(self):
         return 10
+
+
+class RateLimitedAPI:
+    rate_limit = 0.1  # 1 request per 0.1 second
+    last_request_time = 0
+
+    @staticmethod
+    def make_request():
+        current_time = time.time()
+
+        if current_time - RateLimitedAPI.last_request_time < RateLimitedAPI.rate_limit:
+            raise Exception("Rate limit exceeded. Please try again later.")
+
+        # Simulate a successful request
+        RateLimitedAPI.last_request_time = current_time
+        return "Request successful"
+
+
+@registry.register("test-rate-limited")
+class MockRateLimitedEmbeddingFunction(MockTextEmbeddingFunction):
+    def generate_embeddings(self, texts):
+        RateLimitedAPI.make_request()
+        return [self._compute_one_embedding(row) for row in texts]

python/lancedb/context.py

@@ -84,7 +84,9 @@ def contextualize(raw_df: "pd.DataFrame") -> Contextualizer:
     context windows that don't cross document boundaries. In this case, we can
     pass ``document_id`` as the group by.
 
-    >>> contextualize(data).window(4).stride(2).text_col('token').groupby('document_id').to_pandas()
+    >>> (contextualize(data)
+    ...     .window(4).stride(2).text_col('token').groupby('document_id')
+    ...     .to_pandas())
                        token  document_id
     0    The quick brown fox            1
     2  brown fox jumped over            1
@@ -92,18 +94,24 @@ def contextualize(raw_df: "pd.DataFrame") -> Contextualizer:
     6           the lazy dog            1
     9      I love sandwiches            2
 
-    ``min_window_size`` determines the minimum size of the  context windows that are generated
+    ``min_window_size`` determines the minimum size of the context windows
-    This can be used to trim the last few context windows which have size less than
+    that are generated.This can be used to trim the last few context windows
-    ``min_window_size``. By default context windows of size 1 are skipped.
+    which have size less than ``min_window_size``.
+    By default context windows of size 1 are skipped.
 
-    >>> contextualize(data).window(6).stride(3).text_col('token').groupby('document_id').to_pandas()
+    >>> (contextualize(data)
+    ...     .window(6).stride(3).text_col('token').groupby('document_id')
+    ...     .to_pandas())
                                  token  document_id
     0  The quick brown fox jumped over            1
     3     fox jumped over the lazy dog            1
     6                     the lazy dog            1
     9                I love sandwiches            2
 
-    >>> contextualize(data).window(6).stride(3).min_window_size(4).text_col('token').groupby('document_id').to_pandas()
+    >>> (contextualize(data)
+    ...     .window(6).stride(3).min_window_size(4).text_col('token')
+    ...     .groupby('document_id')
+    ...     .to_pandas())
                                  token  document_id
     0  The quick brown fox jumped over            1
     3     fox jumped over the lazy dog            1
@@ -113,7 +121,9 @@ def contextualize(raw_df: "pd.DataFrame") -> Contextualizer:
 
 
 class Contextualizer:
-    """Create context windows from a DataFrame. See [lancedb.context.contextualize][]."""
+    """Create context windows from a DataFrame.
+    See [lancedb.context.contextualize][].
+    """
 
     def __init__(self, raw_df):
         self._text_col = None
@@ -183,7 +193,7 @@ class Contextualizer:
         deprecated_in="0.3.1",
         removed_in="0.4.0",
         current_version=__version__,
-        details="Use the bar function instead",
+        details="Use to_pandas() instead",
     )
     def to_df(self) -> "pd.DataFrame":
         return self.to_pandas()

python/lancedb/db.py

@@ -14,26 +14,39 @@
 from __future__ import annotations
 
 import os
-from abc import ABC, abstractmethod
+from abc import abstractmethod
 from pathlib import Path
-from typing import List, Optional, Union
+from typing import TYPE_CHECKING, Iterable, List, Optional, Union
 
 import pyarrow as pa
+from overrides import EnforceOverrides, override
 from pyarrow import fs
 
-from .common import DATA, URI
-from .embeddings import EmbeddingFunctionConfig
-from .pydantic import LanceModel
 from .table import LanceTable, Table
 from .util import fs_from_uri, get_uri_location, get_uri_scheme
 
+if TYPE_CHECKING:
+    from .common import DATA, URI
+    from .embeddings import EmbeddingFunctionConfig
+    from .pydantic import LanceModel
 
-class DBConnection(ABC):
+
+class DBConnection(EnforceOverrides):
     """An active LanceDB connection interface."""
 
     @abstractmethod
-    def table_names(self) -> list[str]:
+    def table_names(
-        """List all table names in the database."""
+        self, page_token: Optional[str] = None, limit: int = 10
+    ) -> Iterable[str]:
+        """List all table in this database
+
+        Parameters
+        ----------
+        page_token: str, optional
+            The token to use for pagination. If not present, start from the beginning.
+        limit: int, default 10
+            The size of the page to return.
+        """
         pass
 
     @abstractmethod
@@ -45,6 +58,7 @@ class DBConnection(ABC):
         mode: str = "create",
         on_bad_vectors: str = "error",
         fill_value: float = 0.0,
+        embedding_functions: Optional[List[EmbeddingFunctionConfig]] = None,
     ) -> Table:
         """Create a [Table][lancedb.table.Table] in the database.
 
@@ -52,12 +66,24 @@ class DBConnection(ABC):
         ----------
         name: str
             The name of the table.
-        data: list, tuple, dict, pd.DataFrame; optional
+        data: The data to initialize the table, *optional*
-            The data to initialize the table. User must provide at least one of `data` or `schema`.
+            User must provide at least one of `data` or `schema`.
-        schema: pyarrow.Schema or LanceModel; optional
+            Acceptable types are:
-            The schema of the table.
+
+            - dict or list-of-dict
+
+            - pandas.DataFrame
+
+            - pyarrow.Table or pyarrow.RecordBatch
+        schema: The schema of the table, *optional*
+            Acceptable types are:
+
+            - pyarrow.Schema
+
+            - [LanceModel][lancedb.pydantic.LanceModel]
         mode: str; default "create"
-            The mode to use when creating the table. Can be either "create" or "overwrite".
+            The mode to use when creating the table.
+            Can be either "create" or "overwrite".
             By default, if the table already exists, an exception is raised.
             If you want to overwrite the table, use mode="overwrite".
         on_bad_vectors: str, default "error"
@@ -150,7 +176,8 @@ class DBConnection(ABC):
         ...     for i in range(5):
         ...         yield pa.RecordBatch.from_arrays(
         ...             [
-        ...                 pa.array([[3.1, 4.1], [5.9, 26.5]], pa.list_(pa.float32(), 2)),
+        ...                 pa.array([[3.1, 4.1], [5.9, 26.5]],
+        ...                     pa.list_(pa.float32(), 2)),
         ...                 pa.array(["foo", "bar"]),
         ...                 pa.array([10.0, 20.0]),
         ...             ],
@@ -249,12 +276,15 @@ class LanceDBConnection(DBConnection):
     def uri(self) -> str:
         return self._uri
 
-    def table_names(self) -> list[str]:
+    @override
-        """Get the names of all tables in the database.
+    def table_names(
+        self, page_token: Optional[str] = None, limit: int = 10
+    ) -> Iterable[str]:
+        """Get the names of all tables in the database. The names are sorted.
 
         Returns
         -------
-        list of str
+        Iterator of str.
             A list of table names.
         """
         try:
@@ -274,6 +304,7 @@ class LanceDBConnection(DBConnection):
             for file_info in paths
             if file_info.extension == "lance"
         ]
+        tables.sort()
         return tables
 
     def __len__(self) -> int:
@@ -282,6 +313,7 @@ class LanceDBConnection(DBConnection):
     def __contains__(self, name: str) -> bool:
         return name in self.table_names()
 
+    @override
     def create_table(
         self,
         name: str,
@@ -313,6 +345,7 @@ class LanceDBConnection(DBConnection):
         )
         return tbl
 
+    @override
     def open_table(self, name: str) -> LanceTable:
         """Open a table in the database.
 
@@ -327,6 +360,7 @@ class LanceDBConnection(DBConnection):
         """
         return LanceTable.open(self, name)
 
+    @override
     def drop_table(self, name: str, ignore_missing: bool = False):
         """Drop a table from the database.
 
@@ -345,6 +379,7 @@ class LanceDBConnection(DBConnection):
             if not ignore_missing:
                 raise
 
+    @override
     def drop_database(self):
         filesystem, path = fs_from_uri(self.uri)
         filesystem.delete_dir(path)

python/lancedb/embeddings/__init__.py

@@ -11,8 +11,10 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 
+# ruff: noqa: F401
 from .base import EmbeddingFunction, EmbeddingFunctionConfig, TextEmbeddingFunction
 from .cohere import CohereEmbeddingFunction
+from .instructor import InstructorEmbeddingFunction
 from .open_clip import OpenClipEmbeddings
 from .openai import OpenAIEmbeddings
 from .registry import EmbeddingFunctionRegistry, get_registry

python/lancedb/embeddings/base.py

@@ -1,3 +1,15 @@
+#  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 importlib
 from abc import ABC, abstractmethod
 from typing import List, Union
@@ -6,7 +18,7 @@ import numpy as np
 import pyarrow as pa
 from pydantic import BaseModel, Field, PrivateAttr
 
-from .utils import TEXT
+from .utils import TEXT, retry_with_exponential_backoff
 
 
 class EmbeddingFunction(BaseModel, ABC):
@@ -21,6 +33,10 @@ class EmbeddingFunction(BaseModel, ABC):
     3. ndims method which returns the number of dimensions of the vector column
     """
 
+    __slots__ = ("__weakref__",)  # pydantic 1.x compatibility
+    max_retries: int = (
+        7  # Setitng 0 disables retires. Maybe this should not be enabled by default,
+    )
     _ndims: int = PrivateAttr()
 
     @classmethod
@@ -44,6 +60,25 @@ class EmbeddingFunction(BaseModel, ABC):
         """
         pass
 
+    def compute_query_embeddings_with_retry(self, *args, **kwargs) -> List[np.array]:
+        """
+        Compute the embeddings for a given user query with retries
+        """
+        return retry_with_exponential_backoff(
+            self.compute_query_embeddings, max_retries=self.max_retries
+        )(
+            *args,
+            **kwargs,
+        )
+
+    def compute_source_embeddings_with_retry(self, *args, **kwargs) -> List[np.array]:
+        """
+        Compute the embeddings for the source column in the database with retries
+        """
+        return retry_with_exponential_backoff(
+            self.compute_source_embeddings, max_retries=self.max_retries
+        )(*args, **kwargs)
+
     def sanitize_input(self, texts: TEXT) -> Union[List[str], np.ndarray]:
         """
         Sanitize the input to the embedding function.
@@ -103,6 +138,14 @@ class EmbeddingFunction(BaseModel, ABC):
         """
         return Field(json_schema_extra={"vector_column_for": self}, **kwargs)
 
+    def __eq__(self, __value: object) -> bool:
+        if not hasattr(__value, "__dict__"):
+            return False
+        return vars(self) == vars(__value)
+
+    def __hash__(self) -> int:
+        return hash(frozenset(vars(self).items()))
+
 
 class EmbeddingFunctionConfig(BaseModel):
     """

python/lancedb/embeddings/cohere.py

@@ -31,7 +31,8 @@ class CohereEmbeddingFunction(TextEmbeddingFunction):
     Parameters
     ----------
     name: str, default "embed-multilingual-v2.0"
-        The name of the model to use. See the Cohere documentation for a list of available models.
+        The name of the model to use. See the Cohere documentation for
+        a list of available models.
 
     Examples
     --------
@@ -39,7 +40,10 @@ class CohereEmbeddingFunction(TextEmbeddingFunction):
     from lancedb.pydantic import LanceModel, Vector
     from lancedb.embeddings import EmbeddingFunctionRegistry
 
-    cohere = EmbeddingFunctionRegistry.get_instance().get("cohere").create(name="embed-multilingual-v2.0")
+    cohere = EmbeddingFunctionRegistry
+        .get_instance()
+        .get("cohere")
+        .create(name="embed-multilingual-v2.0")
 
     class TextModel(LanceModel):
         text: str = cohere.SourceField()

python/lancedb/embeddings/instructor.py

@@ -0,0 +1,137 @@
+#  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.
+from typing import List
+
+import numpy as np
+
+from .base import TextEmbeddingFunction
+from .registry import register
+from .utils import TEXT, weak_lru
+
+
+@register("instructor")
+class InstructorEmbeddingFunction(TextEmbeddingFunction):
+    """
+    An embedding function that uses the InstructorEmbedding library. Instructor models support multi-task learning, and can be used for a
+    variety of tasks, including text classification, sentence similarity, and document retrieval.
+    If you want to calculate customized embeddings for specific sentences, you may follow the unified template to write instructions:
+        "Represent the `domain` `text_type` for `task_objective`":
+
+        * domain is optional, and it specifies the domain of the text, e.g., science, finance, medicine, etc.
+        * text_type is required, and it specifies the encoding unit, e.g., sentence, document, paragraph, etc.
+        * task_objective is optional, and it specifies the objective of embedding, e.g., retrieve a document, classify the sentence, etc.
+
+    For example, if you want to calculate embeddings for a document, you may write the instruction as follows:
+        "Represent the document for retreival"
+
+    Parameters
+    ----------
+    name: str
+        The name of the model to use. Available models are listed at https://github.com/xlang-ai/instructor-embedding#model-list;
+        The default model is hkunlp/instructor-base
+    batch_size: int, default 32
+        The batch size to use when generating embeddings
+    device: str, default "cpu"
+        The device to use when generating embeddings
+    show_progress_bar: bool, default True
+        Whether to show a progress bar when generating embeddings
+    normalize_embeddings: bool, default True
+        Whether to normalize the embeddings
+    quantize: bool, default False
+        Whether to quantize the model
+    source_instruction: str, default "represent the docuement for retreival"
+        The instruction for the source column
+    query_instruction: str, default "represent the document for retreiving the most similar documents"
+        The instruction for the query
+
+    Examples
+    --------
+    import lancedb
+    from lancedb.pydantic import LanceModel, Vector
+    from lancedb.embeddings import get_registry, InstuctorEmbeddingFunction
+
+    instructor = get_registry().get("instructor").create(
+                                source_instruction="represent the docuement for retreival",
+                                query_instruction="represent the document for retreiving the most similar documents"
+                                )
+
+    class Schema(LanceModel):
+        vector: Vector(instructor.ndims()) = instructor.VectorField()
+        text: str = instructor.SourceField()
+
+    db = lancedb.connect("~/.lancedb")
+    tbl = db.create_table("test", schema=Schema, mode="overwrite")
+
+    texts = [{"text": "Capitalism has been dominant in the Western world since the end of feudalism, but most feel[who?] that..."},
+            {"text": "The disparate impact theory is especially controversial under the Fair Housing Act because the Act..."},
+            {"text": "Disparate impact in United States labor law refers to practices in employment, housing, and other areas that.."}]
+
+    tbl.add(texts)
+
+    """
+
+    name: str = "hkunlp/instructor-base"
+    batch_size: int = 32
+    device: str = "cpu"
+    show_progress_bar: bool = True
+    normalize_embeddings: bool = True
+    quantize: bool = False
+    # convert_to_numpy: bool = True # Hardcoding this as numpy can be ingested directly
+
+    source_instruction: str = "represent the document for retrieval"
+    query_instruction: str = (
+        "represent the document for retrieving the most similar documents"
+    )
+
+    @weak_lru(maxsize=1)
+    def ndims(self):
+        model = self.get_model()
+        return model.encode("foo").shape[0]
+
+    def compute_query_embeddings(self, query: str, *args, **kwargs) -> List[np.array]:
+        return self.generate_embeddings([[self.query_instruction, query]])
+
+    def compute_source_embeddings(self, texts: TEXT, *args, **kwargs) -> List[np.array]:
+        texts = self.sanitize_input(texts)
+        texts_formatted = []
+        for text in texts:
+            texts_formatted.append([self.source_instruction, text])
+        return self.generate_embeddings(texts_formatted)
+
+    def generate_embeddings(self, texts: List) -> List:
+        model = self.get_model()
+        res = model.encode(
+            texts,
+            batch_size=self.batch_size,
+            show_progress_bar=self.show_progress_bar,
+            normalize_embeddings=self.normalize_embeddings,
+        ).tolist()
+        return res
+
+    @weak_lru(maxsize=1)
+    def get_model(self):
+        instructor_embedding = self.safe_import(
+            "InstructorEmbedding", "InstructorEmbedding"
+        )
+        torch = self.safe_import("torch", "torch")
+
+        model = instructor_embedding.INSTRUCTOR(self.name)
+        if self.quantize:
+            if (
+                "qnnpack" in torch.backends.quantized.supported_engines
+            ):  # fix for https://github.com/pytorch/pytorch/issues/29327
+                torch.backends.quantized.engine = "qnnpack"
+            model = torch.quantization.quantize_dynamic(
+                model, {torch.nn.Linear}, dtype=torch.qint8
+            )
+        return model

python/lancedb/embeddings/open_clip.py

@@ -1,3 +1,15 @@
+#  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 concurrent.futures
 import io
 import os

python/lancedb/embeddings/openai.py

@@ -1,3 +1,15 @@
+#  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.
 from typing import List, Union
 
 import numpy as np

python/lancedb/embeddings/sentence_transformers.py

@@ -1,3 +1,15 @@
+#  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.
 from typing import List, Union
 
 import numpy as np
@@ -5,6 +17,7 @@ from cachetools import cached
 
 from .base import TextEmbeddingFunction
 from .registry import register
+from .utils import weak_lru
 
 
 @register("sentence-transformers")
@@ -30,7 +43,7 @@ class SentenceTransformerEmbeddings(TextEmbeddingFunction):
         name and device. This is cached so that the model is only loaded
         once per process.
         """
-        return self.__class__.get_embedding_model(self.name, self.device)
+        return self.get_embedding_model()
 
     def ndims(self):
         if self._ndims is None:
@@ -54,9 +67,8 @@ class SentenceTransformerEmbeddings(TextEmbeddingFunction):
             normalize_embeddings=self.normalize,
         ).tolist()
 
-    @classmethod
+    @weak_lru(maxsize=1)
-    @cached(cache={})
+    def get_embedding_model(self):
-    def get_embedding_model(cls, name, device):
         """
         Get the sentence-transformers embedding model specified by the
         name and device. This is cached so that the model is only loaded
@@ -71,7 +83,7 @@ class SentenceTransformerEmbeddings(TextEmbeddingFunction):
 
         TODO: use lru_cache instead with a reasonable/configurable maxsize
         """
-        sentence_transformers = cls.safe_import(
+        sentence_transformers = self.safe_import(
             "sentence_transformers", "sentence-transformers"
         )
-        return sentence_transformers.SentenceTransformer(name, device=device)
+        return sentence_transformers.SentenceTransformer(self.name, device=self.device)

python/lancedb/embeddings/utils.py

@@ -11,10 +11,14 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 
+import functools
 import math
+import random
 import socket
 import sys
+import time
 import urllib.error
+import weakref
 from typing import Callable, List, Union
 
 import numpy as np
@@ -162,6 +166,99 @@ class FunctionWrapper:
             yield from _chunker(arr)
 
 
+def weak_lru(maxsize=128):
+    """
+    LRU cache that keeps weak references to the objects it caches. Only caches the latest instance of the objects to make sure memory usage
+    is bounded.
+
+    Parameters
+    ----------
+    maxsize : int, default 128
+        The maximum number of objects to cache.
+
+    Returns
+    -------
+    Callable
+        A decorator that can be applied to a method.
+
+    Examples
+    --------
+    >>> class Foo:
+    ...     @weak_lru()
+    ...     def bar(self, x):
+    ...         return x
+    >>> foo = Foo()
+    >>> foo.bar(1)
+    1
+    >>> foo.bar(2)
+    2
+    >>> foo.bar(1)
+    1
+    """
+
+    def wrapper(func):
+        @functools.lru_cache(maxsize)
+        def _func(_self, *args, **kwargs):
+            return func(_self(), *args, **kwargs)
+
+        @functools.wraps(func)
+        def inner(self, *args, **kwargs):
+            return _func(weakref.ref(self), *args, **kwargs)
+
+        return inner
+
+    return wrapper
+
+
+def retry_with_exponential_backoff(
+    func,
+    initial_delay: float = 1,
+    exponential_base: float = 2,
+    jitter: bool = True,
+    max_retries: int = 7,
+    # errors: tuple = (),
+):
+    """Retry a function with exponential backoff.
+
+    Args:
+        func (function): The function to be retried.
+        initial_delay (float): Initial delay in seconds (default is 1).
+        exponential_base (float): The base for exponential backoff (default is 2).
+        jitter (bool): Whether to add jitter to the delay (default is True).
+        max_retries (int): Maximum number of retries (default is 10).
+        errors (tuple): Tuple of specific exceptions to retry on (default is (openai.error.RateLimitError,)).
+
+    Returns:
+        function: The decorated function.
+    """
+
+    def wrapper(*args, **kwargs):
+        num_retries = 0
+        delay = initial_delay
+
+        # Loop until a successful response or max_retries is hit or an exception is raised
+        while True:
+            try:
+                return func(*args, **kwargs)
+
+            # Currently retrying on all exceptions as there is no way to know the format of the error msgs used by different APIs
+            # We'll log the error and say that it is assumed that if this portion errors out, it's due to rate limit but the user
+            # should check the error message to be sure
+            except Exception as e:
+                num_retries += 1
+
+                if num_retries > max_retries:
+                    raise Exception(
+                        f"Maximum number of retries ({max_retries}) exceeded."
+                    )
+
+                delay *= exponential_base * (1 + jitter * random.random())
+                LOGGER.info(f"Retrying in {delay:.2f} seconds due to {e}")
+                time.sleep(delay)
+
+    return wrapper
+
+
 def url_retrieve(url: str):
     """
     Parameters

python/lancedb/query.py

@@ -14,7 +14,7 @@
 from __future__ import annotations
 
 from abc import ABC, abstractmethod
-from typing import List, Literal, Optional, Type, Union
+from typing import TYPE_CHECKING, List, Literal, Optional, Type, Union
 
 import deprecation
 import numpy as np
@@ -23,14 +23,49 @@ import pydantic
 
 from . import __version__
 from .common import VECTOR_COLUMN_NAME
-from .pydantic import LanceModel
 from .util import safe_import_pandas
 
+if TYPE_CHECKING:
+    from .pydantic import LanceModel
+
 pd = safe_import_pandas()
 
 
 class Query(pydantic.BaseModel):
-    """A Query"""
+    """The LanceDB Query
+
+    Attributes
+    ----------
+    vector : List[float]
+        the vector to search for
+    filter : Optional[str]
+        sql filter to refine the query with, optional
+    prefilter : bool
+        if True then apply the filter before vector search
+    k : int
+        top k results to return
+    metric : str
+        the distance metric between a pair of vectors,
+
+        can support L2 (default), Cosine and Dot.
+        [metric definitions][search]
+    columns : Optional[List[str]]
+        which columns to return in the results
+    nprobes : int
+        The number of probes used - optional
+
+        - A higher number makes search more accurate but also slower.
+
+        - See discussion in [Querying an ANN Index][querying-an-ann-index] for
+          tuning advice.
+    refine_factor : Optional[int]
+        Refine the results by reading extra elements and re-ranking them in memory - optional
+
+        - A higher number makes search more accurate but also slower.
+
+        - See discussion in [Querying an ANN Index][querying-an-ann-index] for
+          tuning advice.
+    """
 
     vector_column: str = VECTOR_COLUMN_NAME
 
@@ -61,6 +96,10 @@ class Query(pydantic.BaseModel):
 
 
 class LanceQueryBuilder(ABC):
+    """Build LanceDB query based on specific query type:
+    vector or full text search.
+    """
+
     @classmethod
     def create(
         cls,
@@ -103,7 +142,7 @@ class LanceQueryBuilder(ABC):
             if not isinstance(query, (list, np.ndarray)):
                 conf = table.embedding_functions.get(vector_column_name)
                 if conf is not None:
-                    query = conf.function.compute_query_embeddings(query)[0]
+                    query = conf.function.compute_query_embeddings_with_retry(query)[0]
                 else:
                     msg = f"No embedding function for {vector_column_name}"
                     raise ValueError(msg)
@@ -114,7 +153,7 @@ class LanceQueryBuilder(ABC):
             else:
                 conf = table.embedding_functions.get(vector_column_name)
                 if conf is not None:
-                    query = conf.function.compute_query_embeddings(query)[0]
+                    query = conf.function.compute_query_embeddings_with_retry(query)[0]
                     return query, "vector"
                 else:
                     return query, "fts"
@@ -133,11 +172,11 @@ class LanceQueryBuilder(ABC):
         deprecated_in="0.3.1",
         removed_in="0.4.0",
         current_version=__version__,
-        details="Use the bar function instead",
+        details="Use to_pandas() instead",
     )
     def to_df(self) -> "pd.DataFrame":
         """
-        Deprecated alias for `to_pandas()`. Please use `to_pandas()` instead.
+        *Deprecated alias for `to_pandas()`. Please use `to_pandas()` instead.*
 
         Execute the query and return the results as a pandas DataFrame.
         In addition to the selected columns, LanceDB also returns a vector
@@ -226,13 +265,20 @@ class LanceQueryBuilder(ABC):
         self._columns = columns
         return self
 
-    def where(self, where) -> LanceQueryBuilder:
+    def where(self, where: str, prefilter: bool = False) -> LanceQueryBuilder:
         """Set the where clause.
 
         Parameters
         ----------
         where: str
-            The where clause.
+            The where clause which is a valid SQL where clause. See
+            `Lance filter pushdown <https://lancedb.github.io/lance/read_and_write.html#filter-push-down>`_
+            for valid SQL expressions.
+        prefilter: bool, default False
+            If True, apply the filter before vector search, otherwise the
+            filter is applied on the result of vector search.
+            This feature is **EXPERIMENTAL** and may be removed and modified
+            without warning in the future.
 
         Returns
         -------
@@ -240,13 +286,12 @@ class LanceQueryBuilder(ABC):
             The LanceQueryBuilder object.
         """
         self._where = where
+        self._prefilter = prefilter
         return self
 
 
 class LanceVectorQueryBuilder(LanceQueryBuilder):
     """
-    A builder for nearest neighbor queries for LanceDB.
-
     Examples
     --------
     >>> import lancedb
@@ -302,7 +347,7 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         Higher values will yield better recall (more likely to find vectors if
         they exist) at the expense of latency.
 
-        See discussion in [Querying an ANN Index][../querying-an-ann-index] for
+        See discussion in [Querying an ANN Index][querying-an-ann-index] for
         tuning advice.
 
         Parameters
@@ -369,14 +414,14 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         Parameters
         ----------
         where: str
-            The where clause.
+            The where clause which is a valid SQL where clause. See
+            `Lance filter pushdown <https://lancedb.github.io/lance/read_and_write.html#filter-push-down>`_
+            for valid SQL expressions.
         prefilter: bool, default False
             If True, apply the filter before vector search, otherwise the
             filter is applied on the result of vector search.
             This feature is **EXPERIMENTAL** and may be removed and modified
-            without warning in the future. Currently this is only supported
+            without warning in the future.
-            in OSS and can only be used with a table that does not have an ANN
-            index.
 
         Returns
         -------
@@ -389,6 +434,8 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
 
 
 class LanceFtsQueryBuilder(LanceQueryBuilder):
+    """A builder for full text search for LanceDB."""
+
     def __init__(self, table: "lancedb.table.Table", query: str):
         super().__init__(table)
         self._query = query

python/lancedb/remote/client.py

@@ -13,7 +13,7 @@
 
 
 import functools
-from typing import Any, Callable, Dict, Optional, Union
+from typing import Any, Callable, Dict, Iterable, Optional, Union
 
 import aiohttp
 import attrs
@@ -151,15 +151,14 @@ class RestfulLanceDBClient:
             return await deserialize(resp)
 
     @_check_not_closed
-    async def list_tables(self, limit: int, page_token: str):
+    async def list_tables(
+        self, limit: int, page_token: Optional[str] = None
+    ) -> Iterable[str]:
         """List all tables in the database."""
-        try:
+        if page_token is None:
-            json = await self.get(
+            page_token = ""
-                "/v1/table/", {"limit": limit, "page_token": page_token}
+        json = await self.get("/v1/table/", {"limit": limit, "page_token": page_token})
-            )
+        return json["tables"]
-            return json["tables"]
-        except StopAsyncIteration:
-            return []
 
     @_check_not_closed
     async def query(self, table_name: str, query: VectorQuery) -> VectorQueryResult:

python/lancedb/remote/db.py

@@ -12,17 +12,23 @@
 #  limitations under the License.
 
 import asyncio
+import inspect
+import logging
 import uuid
-from typing import Iterator, Optional
+from typing import Iterable, List, Optional, Union
 from urllib.parse import urlparse
 
 import pyarrow as pa
+from overrides import override
 
 from ..common import DATA
 from ..db import DBConnection
+from ..embeddings import EmbeddingFunctionConfig
+from ..pydantic import LanceModel
 from ..table import Table, _sanitize_data
 from .arrow import to_ipc_binary
 from .client import ARROW_STREAM_CONTENT_TYPE, RestfulLanceDBClient
+from .errors import LanceDBClientError
 
 
 class RemoteDBConnection(DBConnection):
@@ -52,12 +58,18 @@ class RemoteDBConnection(DBConnection):
     def __repr__(self) -> str:
         return f"RemoveConnect(name={self.db_name})"
 
-    def table_names(self, last_token: str, limit=10) -> Iterator[str]:
+    @override
+    def table_names(
+        self, page_token: Optional[str] = None, limit: int = 10
+    ) -> Iterable[str]:
         """List the names of all tables in the database.
+
         Parameters
         ----------
-        last_token: str
+        page_token: str
             The last token to start the new page.
+        limit: int, default 10
+            The maximum number of tables to return for each page.
 
         Returns
         -------
@@ -65,15 +77,16 @@ class RemoteDBConnection(DBConnection):
         """
         while True:
             result = self._loop.run_until_complete(
-                self._client.list_tables(limit, last_token)
+                self._client.list_tables(limit, page_token)
             )
             if len(result) > 0:
-                last_token = result[len(result) - 1]
+                page_token = result[len(result) - 1]
             else:
                 break
             for item in result:
-                yield result
+                yield item
 
+    @override
     def open_table(self, name: str) -> Table:
         """Open a Lance Table in the database.
 
@@ -88,23 +101,142 @@ class RemoteDBConnection(DBConnection):
         """
         from .table import RemoteTable
 
-        # TODO: check if table exists
+        # check if table exists
-
+        try:
+            self._loop.run_until_complete(
+                self._client.post(f"/v1/table/{name}/describe/")
+            )
+        except LanceDBClientError as err:
+            if str(err).startswith("Not found"):
+                logging.error(
+                    f"Table {name} does not exist. "
+                    f"Please first call db.create_table({name}, data)"
+                )
         return RemoteTable(self, name)
 
+    @override
     def create_table(
         self,
         name: str,
         data: DATA = None,
-        schema: pa.Schema = None,
+        schema: Optional[Union[pa.Schema, LanceModel]] = None,
         on_bad_vectors: str = "error",
         fill_value: float = 0.0,
+        embedding_functions: Optional[List[EmbeddingFunctionConfig]] = None,
     ) -> Table:
+        """Create a [Table][lancedb.table.Table] in the database.
+
+        Parameters
+        ----------
+        name: str
+            The name of the table.
+        data: The data to initialize the table, *optional*
+            User must provide at least one of `data` or `schema`.
+            Acceptable types are:
+
+            - dict or list-of-dict
+
+            - pandas.DataFrame
+
+            - pyarrow.Table or pyarrow.RecordBatch
+        schema: The schema of the table, *optional*
+            Acceptable types are:
+
+            - pyarrow.Schema
+
+            - [LanceModel][lancedb.pydantic.LanceModel]
+        on_bad_vectors: str, default "error"
+            What to do if any of the vectors are not the same size or contains NaNs.
+            One of "error", "drop", "fill".
+        fill_value: float
+            The value to use when filling vectors. Only used if on_bad_vectors="fill".
+
+        Returns
+        -------
+        LanceTable
+            A reference to the newly created table.
+
+        !!! note
+
+            The vector index won't be created by default.
+            To create the index, call the `create_index` method on the table.
+
+        Examples
+        --------
+
+        Can create with list of tuples or dictionaries:
+
+        >>> import lancedb
+        >>> db = lancedb.connect("db://test-project-8f45eb")
+        >>> data = [{"vector": [1.1, 1.2], "lat": 45.5, "long": -122.7},
+        ...         {"vector": [0.2, 1.8], "lat": 40.1, "long":  -74.1}]
+        >>> db.create_table("my_table", data)
+        LanceTable(my_table)
+
+        You can also pass a pandas DataFrame:
+
+        >>> import pandas as pd
+        >>> data = pd.DataFrame({
+        ...    "vector": [[1.1, 1.2], [0.2, 1.8]],
+        ...    "lat": [45.5, 40.1],
+        ...    "long": [-122.7, -74.1]
+        ... })
+        >>> db.create_table("table2", data)
+        LanceTable(table2)
+
+        >>> custom_schema = pa.schema([
+        ...   pa.field("vector", pa.list_(pa.float32(), 2)),
+        ...   pa.field("lat", pa.float32()),
+        ...   pa.field("long", pa.float32())
+        ... ])
+        >>> db.create_table("table3", data, schema = custom_schema)
+        LanceTable(table3)
+
+        It is also possible to create an table from `[Iterable[pa.RecordBatch]]`:
+
+        >>> import pyarrow as pa
+        >>> def make_batches():
+        ...     for i in range(5):
+        ...         yield pa.RecordBatch.from_arrays(
+        ...             [
+        ...                 pa.array([[3.1, 4.1], [5.9, 26.5]],
+        ...                     pa.list_(pa.float32(), 2)),
+        ...                 pa.array(["foo", "bar"]),
+        ...                 pa.array([10.0, 20.0]),
+        ...             ],
+        ...             ["vector", "item", "price"],
+        ...         )
+        >>> schema=pa.schema([
+        ...     pa.field("vector", pa.list_(pa.float32(), 2)),
+        ...     pa.field("item", pa.utf8()),
+        ...     pa.field("price", pa.float32()),
+        ... ])
+        >>> db.create_table("table4", make_batches(), schema=schema)
+        LanceTable(table4)
+
+        """
         if data is None and schema is None:
             raise ValueError("Either data or schema must be provided.")
+        if embedding_functions is not None:
+            raise NotImplementedError(
+                "embedding_functions is not supported for remote databases."
+                "Please vote https://github.com/lancedb/lancedb/issues/626 "
+                "for this feature."
+            )
+
+        if inspect.isclass(schema) and issubclass(schema, LanceModel):
+            # convert LanceModel to pyarrow schema
+            # note that it's possible this contains
+            # embedding function metadata already
+            schema = schema.to_arrow_schema()
+
         if data is not None:
             data = _sanitize_data(
-                data, schema, on_bad_vectors=on_bad_vectors, fill_value=fill_value
+                data,
+                schema,
+                metadata=None,
+                on_bad_vectors=on_bad_vectors,
+                fill_value=fill_value,
             )
         else:
             if schema is None:
@@ -126,6 +258,7 @@ class RemoteDBConnection(DBConnection):
         )
         return RemoteTable(self, name)
 
+    @override
     def drop_table(self, name: str):
         """Drop a table from the database.
 

python/lancedb/remote/table.py

@@ -37,34 +37,81 @@ class RemoteTable(Table):
 
     @cached_property
     def schema(self) -> pa.Schema:
-        """Return the schema of the table."""
+        """The [Arrow Schema](https://arrow.apache.org/docs/python/api/datatypes.html#)
+        of this Table
+
+        """
         resp = self._conn._loop.run_until_complete(
             self._conn._client.post(f"/v1/table/{self._name}/describe/")
         )
         schema = json_to_schema(resp["schema"])
         return schema
 
+    @property
+    def version(self) -> int:
+        """Get the current version of the table"""
+        resp = self._conn._loop.run_until_complete(
+            self._conn._client.post(f"/v1/table/{self._name}/describe/")
+        )
+        return resp["version"]
+
     def to_arrow(self) -> pa.Table:
-        """Return the table as an Arrow table."""
+        """to_arrow() is not supported on the LanceDB cloud"""
         raise NotImplementedError("to_arrow() is not supported on the LanceDB cloud")
 
     def to_pandas(self):
-        """Return the table as a Pandas DataFrame.
+        """to_pandas() is not supported on the LanceDB cloud"""
-
-        Intercept `to_arrow()` for better error message.
-        """
         return NotImplementedError("to_pandas() is not supported on the LanceDB cloud")
 
     def create_index(
         self,
         metric="L2",
-        num_partitions=256,
-        num_sub_vectors=96,
         vector_column_name: str = VECTOR_COLUMN_NAME,
-        replace: bool = True,
+        index_cache_size: Optional[int] = None,
-        accelerator: Optional[str] = None,
     ):
-        raise NotImplementedError
+        """Create an index on the table.
+        Currently, the only parameters that matter are
+        the metric and the vector column name.
+
+        Parameters
+        ----------
+        metric : str
+            The metric to use for the index. Default is "L2".
+        vector_column_name : str
+            The name of the vector column. Default is "vector".
+
+        Examples
+        --------
+        >>> import lancedb
+        >>> import uuid
+        >>> from lancedb.schema import vector
+        >>> conn = lancedb.connect("db://...", api_key="...", region="...")
+        >>> table_name = uuid.uuid4().hex
+        >>> schema = pa.schema(
+        ...     [
+        ...             pa.field("id", pa.uint32(), False),
+        ...            pa.field("vector", vector(128), False),
+        ...             pa.field("s", pa.string(), False),
+        ...     ]
+        ... )
+        >>> table = conn.create_table(
+        >>>     table_name,
+        >>>     schema=schema,
+        >>> )
+        >>> table.create_index("L2", "vector")
+        """
+        index_type = "vector"
+
+        data = {
+            "column": vector_column_name,
+            "index_type": index_type,
+            "metric_type": metric,
+            "index_cache_size": index_cache_size,
+        }
+        resp = self._conn._loop.run_until_complete(
+            self._conn._client.post(f"/v1/table/{self._name}/create_index/", data=data)
+        )
+        return resp
 
     def add(
         self,
@@ -73,6 +120,28 @@ class RemoteTable(Table):
         on_bad_vectors: str = "error",
         fill_value: float = 0.0,
     ) -> int:
+        """Add more data to the [Table](Table). It has the same API signature as the OSS version.
+
+        Parameters
+        ----------
+        data: DATA
+            The data to insert into the table. Acceptable types are:
+
+            - dict or list-of-dict
+
+            - pandas.DataFrame
+
+            - pyarrow.Table or pyarrow.RecordBatch
+        mode: str
+            The mode to use when writing the data. Valid values are
+            "append" and "overwrite".
+        on_bad_vectors: str, default "error"
+            What to do if any of the vectors are not the same size or contains NaNs.
+            One of "error", "drop", "fill".
+        fill_value: float, default 0.
+            The value to use when filling vectors. Only used if on_bad_vectors="fill".
+
+        """
         data = _sanitize_data(
             data,
             self.schema,
@@ -96,16 +165,112 @@ class RemoteTable(Table):
     def search(
         self, query: Union[VEC, str], vector_column_name: str = VECTOR_COLUMN_NAME
     ) -> LanceVectorQueryBuilder:
+        """Create a search query to find the nearest neighbors
+        of the given query vector. We currently support [vector search][search]
+
+        All query options are defined in [Query][lancedb.query.Query].
+
+        Examples
+        --------
+        >>> import lancedb
+        >>> db = lancedb.connect("db://...", api_key="...", region="...")
+        >>> data = [
+        ...    {"original_width": 100, "caption": "bar", "vector": [0.1, 2.3, 4.5]},
+        ...    {"original_width": 2000, "caption": "foo",  "vector": [0.5, 3.4, 1.3]},
+        ...    {"original_width": 3000, "caption": "test", "vector": [0.3, 6.2, 2.6]}
+        ... ]
+        >>> table = db.create_table("my_table", data)
+        >>> query = [0.4, 1.4, 2.4]
+        >>> (table.search(query, vector_column_name="vector")
+        ...     .where("original_width > 1000", prefilter=True)
+        ...     .select(["caption", "original_width"])
+        ...     .limit(2)
+        ...     .to_pandas())
+          caption  original_width           vector  _distance
+        0     foo            2000  [0.5, 3.4, 1.3]   5.220000
+        1    test            3000  [0.3, 6.2, 2.6]  23.089996
+
+        Parameters
+        ----------
+        query: list/np.ndarray/str/PIL.Image.Image, default None
+            The targetted vector to search for.
+
+            - *default None*.
+            Acceptable types are: list, np.ndarray, PIL.Image.Image
+
+            - If None then the select/where/limit clauses are applied to filter
+            the table
+        vector_column_name: str
+            The name of the vector column to search.
+            *default "vector"*
+
+        Returns
+        -------
+        LanceQueryBuilder
+            A query builder object representing the query.
+            Once executed, the query returns
+
+            - selected columns
+
+            - the vector
+
+            - and also the "_distance" column which is the distance between the query
+            vector and the returned vector.
+        """
         return LanceVectorQueryBuilder(self, query, vector_column_name)
 
     def _execute_query(self, query: Query) -> pa.Table:
-        if query.prefilter:
-            raise NotImplementedError("Cloud support for prefiltering is coming soon")
         result = self._conn._client.query(self._name, query)
         return self._conn._loop.run_until_complete(result).to_arrow()
 
     def delete(self, predicate: str):
-        """Delete rows from the table."""
+        """Delete rows from the table.
+
+        This can be used to delete a single row, many rows, all rows, or
+        sometimes no rows (if your predicate matches nothing).
+
+        Parameters
+        ----------
+        predicate: str
+            The SQL where clause to use when deleting rows.
+
+            - For example, 'x = 2' or 'x IN (1, 2, 3)'.
+
+            The filter must not be empty, or it will error.
+
+        Examples
+        --------
+        >>> import lancedb
+        >>> data = [
+        ...    {"x": 1, "vector": [1, 2]},
+        ...    {"x": 2, "vector": [3, 4]},
+        ...    {"x": 3, "vector": [5, 6]}
+        ... ]
+        >>> db = lancedb.connect("db://...", api_key="...", region="...")
+        >>> table = db.create_table("my_table", data)
+        >>> table.search([10,10]).to_pandas()
+           x      vector  _distance
+        0  3  [5.0, 6.0]       41.0
+        1  2  [3.0, 4.0]       85.0
+        2  1  [1.0, 2.0]      145.0
+        >>> table.delete("x = 2")
+        >>> table.search([10,10]).to_pandas()
+           x      vector  _distance
+        0  3  [5.0, 6.0]       41.0
+        1  1  [1.0, 2.0]      145.0
+
+        If you have a list of values to delete, you can combine them into a
+        stringified list and use the `IN` operator:
+
+        >>> to_remove = [1, 3]
+        >>> to_remove = ", ".join([str(v) for v in to_remove])
+        >>> to_remove
+        '1, 3'
+        >>> table.delete(f"x IN ({to_remove})")
+        >>> table.search([10,10]).to_pandas()
+           x      vector  _distance
+        0  2  [3.0, 4.0]       85.0
+        """
         payload = {"predicate": predicate}
         self._conn._loop.run_until_complete(
             self._conn._client.post(f"/v1/table/{self._name}/delete/", data=payload)

python/lancedb/table.py

@@ -16,16 +16,14 @@ from __future__ import annotations
 import inspect
 import os
 from abc import ABC, abstractmethod
-from datetime import timedelta
 from functools import cached_property
-from typing import Any, Iterable, List, Optional, Union
+from typing import TYPE_CHECKING, Any, Iterable, List, Optional, Union
 
 import lance
 import numpy as np
 import pyarrow as pa
 import pyarrow.compute as pc
 from lance import LanceDataset
-from lance.dataset import CleanupStats, ReaderLike
 from lance.vector import vec_to_table
 
 from .common import DATA, VEC, VECTOR_COLUMN_NAME
@@ -35,6 +33,12 @@ from .query import LanceQueryBuilder, Query
 from .util import fs_from_uri, safe_import_pandas
 from .utils.events import register_event
 
+if TYPE_CHECKING:
+    from datetime import timedelta
+
+    from lance.dataset import CleanupStats, ReaderLike
+
+
 pd = safe_import_pandas()
 
 
@@ -86,7 +90,9 @@ def _append_vector_col(data: pa.Table, metadata: dict, schema: Optional[pa.Schem
     for vector_column, conf in functions.items():
         func = conf.function
         if vector_column not in data.column_names:
-            col_data = func.compute_source_embeddings(data[conf.source_column])
+            col_data = func.compute_source_embeddings_with_retry(
+                data[conf.source_column]
+            )
             if schema is not None:
                 dtype = schema.field(vector_column).type
             else:
@@ -149,13 +155,13 @@ class Table(ABC):
     @property
     @abstractmethod
     def schema(self) -> pa.Schema:
-        """The [Arrow Schema](https://arrow.apache.org/docs/python/api/datatypes.html#) of
+        """The [Arrow Schema](https://arrow.apache.org/docs/python/api/datatypes.html#)
-        this Table
+        of this Table
 
         """
         raise NotImplementedError
 
-    def to_pandas(self):
+    def to_pandas(self) -> "pd.DataFrame":
         """Return the table as a pandas DataFrame.
 
         Returns
@@ -182,6 +188,7 @@ class Table(ABC):
         vector_column_name: str = VECTOR_COLUMN_NAME,
         replace: bool = True,
         accelerator: Optional[str] = None,
+        index_cache_size: Optional[int] = None,
     ):
         """Create an index on the table.
 
@@ -191,20 +198,23 @@ class Table(ABC):
             The distance metric to use when creating the index.
             Valid values are "L2", "cosine", or "dot".
             L2 is euclidean distance.
-        num_partitions: int
+        num_partitions: int, default 256
             The number of IVF partitions to use when creating the index.
             Default is 256.
-        num_sub_vectors: int
+        num_sub_vectors: int, default 96
             The number of PQ sub-vectors to use when creating the index.
             Default is 96.
         vector_column_name: str, default "vector"
             The vector column name to create the index.
         replace: bool, default True
-            If True, replace the existing index if it exists.
+            - If True, replace the existing index if it exists.
-            If False, raise an error if duplicate index exists.
+
+            - If False, raise an error if duplicate index exists.
         accelerator: str, default None
             If set, use the given accelerator to create the index.
             Only support "cuda" for now.
+        index_cache_size : int, optional
+            The size of the index cache in number of entries. Default value is 256.
         """
         raise NotImplementedError
 
@@ -220,8 +230,14 @@ class Table(ABC):
 
         Parameters
         ----------
-        data: list-of-dict, dict, pd.DataFrame
+        data: DATA
-            The data to insert into the table.
+            The data to insert into the table. Acceptable types are:
+
+            - dict or list-of-dict
+
+            - pandas.DataFrame
+
+            - pyarrow.Table or pyarrow.RecordBatch
         mode: str
             The mode to use when writing the data. Valid values are
             "append" and "overwrite".
@@ -242,31 +258,70 @@ class Table(ABC):
         query_type: str = "auto",
     ) -> LanceQueryBuilder:
         """Create a search query to find the nearest neighbors
-        of the given query vector.
+        of the given query vector. We currently support [vector search][search]
+        and [full-text search][experimental-full-text-search].
+
+        All query options are defined in [Query][lancedb.query.Query].
+
+        Examples
+        --------
+        >>> import lancedb
+        >>> db = lancedb.connect("./.lancedb")
+        >>> data = [
+        ...    {"original_width": 100, "caption": "bar", "vector": [0.1, 2.3, 4.5]},
+        ...    {"original_width": 2000, "caption": "foo",  "vector": [0.5, 3.4, 1.3]},
+        ...    {"original_width": 3000, "caption": "test", "vector": [0.3, 6.2, 2.6]}
+        ... ]
+        >>> table = db.create_table("my_table", data)
+        >>> query = [0.4, 1.4, 2.4]
+        >>> (table.search(query, vector_column_name="vector")
+        ...     .where("original_width > 1000", prefilter=True)
+        ...     .select(["caption", "original_width"])
+        ...     .limit(2)
+        ...     .to_pandas())
+          caption  original_width           vector  _distance
+        0     foo            2000  [0.5, 3.4, 1.3]   5.220000
+        1    test            3000  [0.3, 6.2, 2.6]  23.089996
 
         Parameters
         ----------
-        query: str, list, np.ndarray, PIL.Image.Image, default None
+        query: list/np.ndarray/str/PIL.Image.Image, default None
-            The query to search for. If None then
+            The targetted vector to search for.
-            the select/where/limit clauses are applied to filter
+
+            - *default None*.
+            Acceptable types are: list, np.ndarray, PIL.Image.Image
+
+            - If None then the select/where/limit clauses are applied to filter
             the table
-        vector_column_name: str, default "vector"
+        vector_column_name: str
             The name of the vector column to search.
-        query_type: str, default "auto"
+            *default "vector"*
-            "vector", "fts", or "auto"
+        query_type: str
-            If "auto" then the query type is inferred from the query;
+            *default "auto"*.
-            If `query` is a list/np.ndarray then the query type is "vector";
+            Acceptable types are: "vector", "fts", or "auto"
-            If `query` is a PIL.Image.Image then either do vector search
+
-            or raise an error if no corresponding embedding function is found.
+            - If "auto" then the query type is inferred from the query;
-            If `query` is a string, then the query type is "vector" if the
+
+                - If `query` is a list/np.ndarray then the query type is
+                "vector";
+
+                - If `query` is a PIL.Image.Image then either do vector search,
+                or raise an error if no corresponding embedding function is found.
+
+            - If `query` is a string, then the query type is "vector" if the
             table has embedding functions else the query type is "fts"
 
         Returns
         -------
         LanceQueryBuilder
             A query builder object representing the query.
-            Once executed, the query returns selected columns, the vector,
+            Once executed, the query returns
-            and also the "_distance" column which is the distance between the query
+
+            - selected columns
+
+            - the vector
+
+            - and also the "_distance" column which is the distance between the query
             vector and the returned vector.
         """
         raise NotImplementedError
@@ -285,14 +340,19 @@ class Table(ABC):
         Parameters
         ----------
         where: str
-            The SQL where clause to use when deleting rows. For example, 'x = 2'
+            The SQL where clause to use when deleting rows.
-            or 'x IN (1, 2, 3)'. The filter must not be empty, or it will error.
+
+            - For example, 'x = 2' or 'x IN (1, 2, 3)'.
+
+            The filter must not be empty, or it will error.
 
         Examples
         --------
         >>> import lancedb
         >>> data = [
-        ...   {"x": 1, "vector": [1, 2]}, {"x": 2, "vector": [3, 4]}, {"x": 3, "vector": [5, 6]}
+        ...    {"x": 1, "vector": [1, 2]},
+        ...    {"x": 2, "vector": [3, 4]},
+        ...    {"x": 3, "vector": [5, 6]}
         ... ]
         >>> db = lancedb.connect("./.lancedb")
         >>> table = db.create_table("my_table", data)
@@ -377,7 +437,8 @@ class LanceTable(Table):
         --------
         >>> import lancedb
         >>> db = lancedb.connect("./.lancedb")
-        >>> table = db.create_table("my_table", [{"vector": [1.1, 0.9], "type": "vector"}])
+        >>> table = db.create_table("my_table",
+        ...    [{"vector": [1.1, 0.9], "type": "vector"}])
         >>> table.version
         2
         >>> table.to_pandas()
@@ -424,7 +485,8 @@ class LanceTable(Table):
         --------
         >>> import lancedb
         >>> db = lancedb.connect("./.lancedb")
-        >>> table = db.create_table("my_table", [{"vector": [1.1, 0.9], "type": "vector"}])
+        >>> table = db.create_table("my_table", [
+        ...     {"vector": [1.1, 0.9], "type": "vector"}])
         >>> table.version
         2
         >>> table.to_pandas()
@@ -497,6 +559,7 @@ class LanceTable(Table):
         vector_column_name=VECTOR_COLUMN_NAME,
         replace: bool = True,
         accelerator: Optional[str] = None,
+        index_cache_size: Optional[int] = None,
     ):
         """Create an index on the table."""
         self._dataset.create_index(
@@ -507,6 +570,7 @@ class LanceTable(Table):
             num_sub_vectors=num_sub_vectors,
             replace=replace,
             accelerator=accelerator,
+            index_cache_size=index_cache_size,
         )
         self._reset_dataset()
         register_event("create_index")
@@ -669,14 +733,39 @@ class LanceTable(Table):
         query_type: str = "auto",
     ) -> LanceQueryBuilder:
         """Create a search query to find the nearest neighbors
-        of the given query vector.
+        of the given query vector. We currently support [vector search][search]
+        and [full-text search][search].
+
+        Examples
+        --------
+        >>> import lancedb
+        >>> db = lancedb.connect("./.lancedb")
+        >>> data = [
+        ...    {"original_width": 100, "caption": "bar", "vector": [0.1, 2.3, 4.5]},
+        ...    {"original_width": 2000, "caption": "foo",  "vector": [0.5, 3.4, 1.3]},
+        ...    {"original_width": 3000, "caption": "test", "vector": [0.3, 6.2, 2.6]}
+        ... ]
+        >>> table = db.create_table("my_table", data)
+        >>> query = [0.4, 1.4, 2.4]
+        >>> (table.search(query, vector_column_name="vector")
+        ...     .where("original_width > 1000", prefilter=True)
+        ...     .select(["caption", "original_width"])
+        ...     .limit(2)
+        ...     .to_pandas())
+          caption  original_width           vector  _distance
+        0     foo            2000  [0.5, 3.4, 1.3]   5.220000
+        1    test            3000  [0.3, 6.2, 2.6]  23.089996
 
         Parameters
         ----------
-        query: str, list, np.ndarray, a PIL Image or None
+        query: list/np.ndarray/str/PIL.Image.Image, default None
-            The query to search for. If None then
+            The targetted vector to search for.
-            the select/where/limit clauses are applied to filter
+
-            the table
+            - *default None*.
+            Acceptable types are: list, np.ndarray, PIL.Image.Image
+
+            - If None then the select/[where][sql]/limit clauses are applied
+            to filter the table
         vector_column_name: str, default "vector"
             The name of the vector column to search.
         query_type: str, default "auto"
@@ -685,7 +774,7 @@ class LanceTable(Table):
             If `query` is a list/np.ndarray then the query type is "vector";
             If `query` is a PIL.Image.Image then either do vector search
             or raise an error if no corresponding embedding function is found.
-            If the query is a string, then the query type is "vector" if the
+            If the `query` is a string, then the query type is "vector" if the
             table has embedding functions, else the query type is "fts"
 
         Returns
@@ -720,7 +809,9 @@ class LanceTable(Table):
         --------
         >>> import lancedb
         >>> data = [
-        ...   {"x": 1, "vector": [1, 2]}, {"x": 2, "vector": [3, 4]}, {"x": 3, "vector": [5, 6]}
+        ...    {"x": 1, "vector": [1, 2]},
+        ...    {"x": 2, "vector": [3, 4]},
+        ...    {"x": 3, "vector": [5, 6]}
         ... ]
         >>> db = lancedb.connect("./.lancedb")
         >>> table = db.create_table("my_table", data)
@@ -740,7 +831,8 @@ class LanceTable(Table):
             The data to insert into the table.
             At least one of `data` or `schema` must be provided.
         schema: pa.Schema or LanceModel, optional
-            The schema of the table. If not provided, the schema is inferred from the data.
+            The schema of the table. If not provided,
+            the schema is inferred from the data.
             At least one of `data` or `schema` must be provided.
         mode: str, default "create"
             The mode to use when writing the data. Valid values are
@@ -811,7 +903,8 @@ class LanceTable(Table):
         file_info = fs.get_file_info(path)
         if file_info.type != pa.fs.FileType.Directory:
             raise FileNotFoundError(
-                f"Table {name} does not exist. Please first call db.create_table({name}, data)"
+                f"Table {name} does not exist."
+                f"Please first call db.create_table({name}, data)"
             )
         return tbl
 
@@ -838,7 +931,9 @@ class LanceTable(Table):
         --------
         >>> import lancedb
         >>> data = [
-        ...   {"x": 1, "vector": [1, 2]}, {"x": 2, "vector": [3, 4]}, {"x": 3, "vector": [5, 6]}
+        ...    {"x": 1, "vector": [1, 2]},
+        ...    {"x": 2, "vector": [3, 4]},
+        ...    {"x": 3, "vector": [5, 6]}
         ... ]
         >>> db = lancedb.connect("./.lancedb")
         >>> table = db.create_table("my_table", data)
@@ -872,12 +967,6 @@ class LanceTable(Table):
 
     def _execute_query(self, query: Query) -> pa.Table:
         ds = self.to_lance()
-        if query.prefilter:
-            for idx in ds.list_indices():
-                if query.vector_column in idx["fields"]:
-                    raise NotImplementedError(
-                        "Prefiltering for indexed vector column is coming soon."
-                    )
         return ds.to_table(
             columns=query.columns,
             filter=query.filter,
@@ -1019,7 +1108,8 @@ def _sanitize_vector_column(
     # ChunkedArray is annoying to work with, so we combine chunks here
     vec_arr = data[vector_column_name].combine_chunks()
     if pa.types.is_list(data[vector_column_name].type):
-        # if it's a variable size list array we make sure the dimensions are all the same
+        # if it's a variable size list array,
+        # we make sure the dimensions are all the same
         has_jagged_ndims = len(vec_arr.values) % len(data) != 0
         if has_jagged_ndims:
             data = _sanitize_jagged(

python/lancedb/utils/sentry_log.py

@@ -63,7 +63,8 @@ def set_sentry():
         """
         if "exc_info" in hint:
             exc_type, exc_value, tb = hint["exc_info"]
-            if "out of memory" in str(exc_value).lower():
+            ignored_errors = ["out of memory", "no space left on device", "testing"]
+            if any(error in str(exc_value).lower() for error in ignored_errors):
                 return None
 
         if is_git_dir():
@@ -97,7 +98,7 @@ def set_sentry():
             dsn="https://c63ef8c64e05d1aa1a96513361f3ca2f@o4505950840946688.ingest.sentry.io/4505950933614592",
             debug=False,
             include_local_variables=False,
-            traces_sample_rate=1.0,
+            traces_sample_rate=0.5,
             environment="production",  # 'dev' or 'production'
             before_send=before_send,
             ignore_errors=[KeyboardInterrupt, FileNotFoundError, bdb.BdbQuit],

python/pyproject.toml

@@ -1,9 +1,9 @@
 [project]
 name = "lancedb"
-version = "0.3.2"
+version = "0.3.4"
 dependencies = [
     "deprecation",
-    "pylance==0.8.7",
+    "pylance==0.8.17",
     "ratelimiter~=1.0",
     "retry>=0.9.2",
     "tqdm>=4.1.0",
@@ -14,7 +14,8 @@ dependencies = [
     "cachetools",
     "pyyaml>=6.0",
     "click>=8.1.7",
-    "requests>=2.31.0"
+    "requests>=2.31.0",
+    "overrides>=0.7"
 ]
 description = "lancedb"
 authors = [{ name = "LanceDB Devs", email = "dev@lancedb.com" }]
@@ -52,7 +53,7 @@ tests = ["pandas>=1.4", "pytest", "pytest-mock", "pytest-asyncio", "requests"]
 dev = ["ruff", "pre-commit", "black"]
 docs = ["mkdocs", "mkdocs-jupyter", "mkdocs-material", "mkdocstrings[python]"]
 clip = ["torch", "pillow", "open-clip"]
-embeddings = ["openai", "sentence-transformers", "torch", "pillow", "open-clip-torch", "cohere"]
+embeddings = ["openai", "sentence-transformers", "torch", "pillow", "open-clip-torch", "cohere", "InstructorEmbedding"]
 
 [project.scripts]
 lancedb = "lancedb.cli.cli:cli"
@@ -64,6 +65,9 @@ build-backend = "setuptools.build_meta"
 [tool.isort]
 profile = "black"
 
+[tool.ruff]
+select = ["F", "E", "W", "I", "G", "TCH", "PERF"]
+
 [tool.pytest.ini_options]
 addopts = "--strict-markers"
 markers = [

python/tests/test_db.py

@@ -129,7 +129,7 @@ def test_ingest_iterator(tmp_path):
                 [
                     PydanticSchema(vector=[3.1, 4.1], item="foo", price=10.0),
                     PydanticSchema(vector=[5.9, 26.5], item="bar", price=20.0),
-                ]
+                ],
                 # TODO: test pydict separately. it is unique column number and names contraint
             ]
 
@@ -150,6 +150,21 @@ def test_ingest_iterator(tmp_path):
     run_tests(PydanticSchema)
 
 
+def test_table_names(tmp_path):
+    db = lancedb.connect(tmp_path)
+    data = pd.DataFrame(
+        {
+            "vector": [[3.1, 4.1], [5.9, 26.5]],
+            "item": ["foo", "bar"],
+            "price": [10.0, 20.0],
+        }
+    )
+    db.create_table("test2", data=data)
+    db.create_table("test1", data=data)
+    db.create_table("test3", data=data)
+    assert db.table_names() == ["test1", "test2", "test3"]
+
+
 def test_create_mode(tmp_path):
     db = lancedb.connect(tmp_path)
     data = pd.DataFrame(
@@ -286,4 +301,29 @@ def test_replace_index(tmp_path):
         num_partitions=2,
         num_sub_vectors=4,
         replace=True,
+        index_cache_size=10,
     )
+
+
+def test_prefilter_with_index(tmp_path):
+    db = lancedb.connect(uri=tmp_path)
+    data = [
+        {"vector": np.random.rand(128), "item": "foo", "price": float(i)}
+        for i in range(1000)
+    ]
+    sample_key = data[100]["vector"]
+    table = db.create_table(
+        "test",
+        data,
+    )
+    table.create_index(
+        num_partitions=2,
+        num_sub_vectors=4,
+    )
+    table = (
+        table.search(sample_key)
+        .where("price == 500", prefilter=True)
+        .limit(5)
+        .to_arrow()
+    )
+    assert table.num_rows == 1

python/tests/test_embeddings.py

@@ -15,13 +15,16 @@ import sys
 import lance
 import numpy as np
 import pyarrow as pa
+import pytest
 
-from lancedb.conftest import MockTextEmbeddingFunction
+import lancedb
+from lancedb.conftest import MockRateLimitedEmbeddingFunction, MockTextEmbeddingFunction
 from lancedb.embeddings import (
     EmbeddingFunctionConfig,
     EmbeddingFunctionRegistry,
     with_embeddings,
 )
+from lancedb.pydantic import LanceModel, Vector
 
 
 def mock_embed_func(input_data):
@@ -83,3 +86,29 @@ def test_embedding_function(tmp_path):
     expected = func.compute_query_embeddings("hello world")
 
     assert np.allclose(actual, expected)
+
+
+def test_embedding_function_rate_limit(tmp_path):
+    def _get_schema_from_model(model):
+        class Schema(LanceModel):
+            text: str = model.SourceField()
+            vector: Vector(model.ndims()) = model.VectorField()
+
+        return Schema
+
+    db = lancedb.connect(tmp_path)
+    registry = EmbeddingFunctionRegistry.get_instance()
+    model = registry.get("test-rate-limited").create(max_retries=0)
+    schema = _get_schema_from_model(model)
+    table = db.create_table("test", schema=schema, mode="overwrite")
+    table.add([{"text": "hello world"}])
+    with pytest.raises(Exception):
+        table.add([{"text": "hello world"}])
+    assert len(table) == 1
+
+    model = registry.get("test-rate-limited").create()
+    schema = _get_schema_from_model(model)
+    table = db.create_table("test", schema=schema, mode="overwrite")
+    table.add([{"text": "hello world"}])
+    table.add([{"text": "hello world"}])
+    assert len(table) == 2

python/tests/test_embeddings_slow.py

@@ -32,8 +32,8 @@ from lancedb.pydantic import LanceModel, Vector
 def test_sentence_transformer(alias, tmp_path):
     db = lancedb.connect(tmp_path)
     registry = get_registry()
-    func = registry.get(alias).create()
+    func = registry.get(alias).create(max_retries=0)
-    func2 = registry.get(alias).create()
+    func2 = registry.get(alias).create(max_retries=0)
 
     class Words(LanceModel):
         text: str = func.SourceField()
@@ -150,7 +150,11 @@ def test_openclip(tmp_path):
     os.environ.get("COHERE_API_KEY") is None, reason="COHERE_API_KEY not set"
 )  # also skip if cohere not installed
 def test_cohere_embedding_function():
-    cohere = get_registry().get("cohere").create(name="embed-multilingual-v2.0")
+    cohere = (
+        get_registry()
+        .get("cohere")
+        .create(name="embed-multilingual-v2.0", max_retries=0)
+    )
 
     class TextModel(LanceModel):
         text: str = cohere.SourceField()
@@ -162,3 +166,19 @@ def test_cohere_embedding_function():
 
     tbl.add(df)
     assert len(tbl.to_pandas()["vector"][0]) == cohere.ndims()
+
+
+@pytest.mark.slow
+def test_instructor_embedding(tmp_path):
+    model = get_registry().get("instructor").create()
+
+    class TextModel(LanceModel):
+        text: str = model.SourceField()
+        vector: Vector(model.ndims()) = model.VectorField()
+
+    df = pd.DataFrame({"text": ["hello world", "goodbye world"]})
+    db = lancedb.connect(tmp_path)
+    tbl = db.create_table("test", schema=TextModel, mode="overwrite")
+
+    tbl.add(df)
+    assert len(tbl.to_pandas()["vector"][0]) == model.ndims()

python/tests/test_remote_db.py

@@ -26,6 +26,9 @@ class FakeLanceDBClient:
         t = pa.schema([]).empty_table()
         return VectorQueryResult(t)
 
+    async def post(self, path: str):
+        pass
+
 
 def test_remote_db():
     conn = lancedb.connect("db://client-will-be-injected", api_key="fake")

python/tests/test_table.py

@@ -213,6 +213,7 @@ def test_create_index_method():
                 num_sub_vectors=96,
                 vector_column_name="vector",
                 replace=True,
+                index_cache_size=256,
             )
 
             # Check that the _dataset.create_index method was called
@@ -225,6 +226,7 @@ def test_create_index_method():
                 num_sub_vectors=96,
                 replace=True,
                 accelerator=None,
+                index_cache_size=256,
             )
 
 

rust/ffi/node/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "vectordb-node"
-version = "0.3.3"
+version = "0.3.9"
 description = "Serverless, low-latency vector database for AI applications"
 license = "Apache-2.0"
 edition = "2018"
@@ -19,6 +19,7 @@ once_cell = "1"
 futures = "0.3"
 half = { workspace = true }
 lance = { workspace = true }
+lance-index = { workspace = true }
 lance-linalg = { workspace = true }
 vectordb = { path = "../../vectordb" }
 tokio = { version = "1.23", features = ["rt-multi-thread"] }

rust/ffi/node/src/index/vector.rs

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use lance::index::vector::{ivf::IvfBuildParams, pq::PQBuildParams};
+use lance_index::vector::{ivf::IvfBuildParams, pq::PQBuildParams};
 use lance_linalg::distance::MetricType;
 use neon::context::FunctionContext;
 use neon::prelude::*;
@@ -70,7 +70,6 @@ fn get_index_params_builder(
                 .map(|mt| {
                     let metric_type = mt.unwrap();
                     index_builder.metric_type(metric_type);
-                    pq_params.metric_type = metric_type;
                 });
 
             let num_partitions = obj.get_opt_usize(cx, "num_partitions")?;

rust/ffi/node/src/lib.rs

@@ -239,6 +239,8 @@ fn main(mut cx: ModuleContext) -> NeonResult<()> {
     cx.export_function("tableDelete", JsTable::js_delete)?;
     cx.export_function("tableCleanupOldVersions", JsTable::js_cleanup)?;
     cx.export_function("tableCompactFiles", JsTable::js_compact)?;
+    cx.export_function("tableListIndices", JsTable::js_list_indices)?;
+    cx.export_function("tableIndexStats", JsTable::js_index_stats)?;
     cx.export_function(
         "tableCreateVectorIndex",
         index::vector::table_create_vector_index,

rust/ffi/node/src/query.rs

@@ -48,6 +48,8 @@ impl JsQuery {
             .map(|s| s.value(&mut cx))
             .map(|s| MetricType::try_from(s.as_str()).unwrap());
 
+        let prefilter = query_obj.get::<JsBoolean, _, _>(&mut cx, "_prefilter")?.value(&mut cx);
+
         let is_electron = cx
             .argument::<JsBoolean>(1)
             .or_throw(&mut cx)?
@@ -69,7 +71,8 @@ impl JsQuery {
                 .nprobes(nprobes)
                 .filter(filter)
                 .metric_type(metric_type)
-                .select(select);
+                .select(select)
+                .prefilter(prefilter);
             let record_batch_stream = builder.execute();
             let results = record_batch_stream
                 .and_then(|stream| {

rust/ffi/node/src/table.rs

@@ -247,7 +247,7 @@ impl JsTable {
         }
 
         rt.spawn(async move {
-            let stats = table.compact_files(options).await;
+            let stats = table.compact_files(options, None).await;
 
             deferred.settle_with(&channel, move |mut cx| {
                 let stats = stats.or_throw(&mut cx)?;
@@ -276,4 +276,91 @@ impl JsTable {
         });
         Ok(promise)
     }
+
+    pub(crate) fn js_list_indices(mut cx: FunctionContext) -> JsResult<JsPromise> {
+        let js_table = cx.this().downcast_or_throw::<JsBox<JsTable>, _>(&mut cx)?;
+        let rt = runtime(&mut cx)?;
+        let (deferred, promise) = cx.promise();
+        // let predicate = cx.argument::<JsString>(0)?.value(&mut cx);
+        let channel = cx.channel();
+        let table = js_table.table.clone();
+
+        rt.spawn(async move {
+            let indices = table.load_indices().await;
+
+            deferred.settle_with(&channel, move |mut cx| {
+                let indices = indices.or_throw(&mut cx)?;
+
+                let output = JsArray::new(&mut cx, indices.len() as u32);
+                for (i, index) in indices.iter().enumerate() {
+                    let js_index = JsObject::new(&mut cx);
+                    let index_name = cx.string(index.index_name.clone());
+                    js_index.set(&mut cx, "name", index_name)?;
+
+                    let index_uuid = cx.string(index.index_uuid.clone());
+                    js_index.set(&mut cx, "uuid", index_uuid)?;
+
+                    let js_index_columns = JsArray::new(&mut cx, index.columns.len() as u32);
+                    for (j, column) in index.columns.iter().enumerate() {
+                        let js_column = cx.string(column.clone());
+                        js_index_columns.set(&mut cx, j as u32, js_column)?;
+                    }
+                    js_index.set(&mut cx, "columns", js_index_columns)?;
+
+                    output.set(&mut cx, i as u32, js_index)?;
+                }
+
+                Ok(output)
+            })
+        });
+        Ok(promise)
+    }
+
+    pub(crate) fn js_index_stats(mut cx: FunctionContext) -> JsResult<JsPromise> {
+        let js_table = cx.this().downcast_or_throw::<JsBox<JsTable>, _>(&mut cx)?;
+        let rt = runtime(&mut cx)?;
+        let (deferred, promise) = cx.promise();
+        let index_uuid = cx.argument::<JsString>(0)?.value(&mut cx);
+        let channel = cx.channel();
+        let table = js_table.table.clone();
+
+        rt.spawn(async move {
+            let load_stats = futures::try_join!(
+                table.count_indexed_rows(&index_uuid),
+                table.count_unindexed_rows(&index_uuid)
+            );
+
+            deferred.settle_with(&channel, move |mut cx| {
+                let (indexed_rows, unindexed_rows) = load_stats.or_throw(&mut cx)?;
+
+                let output = JsObject::new(&mut cx);
+
+                match indexed_rows {
+                    Some(x) => {
+                        let i = cx.number(x as f64);
+                        output.set(&mut cx, "numIndexedRows", i)?;
+                    }
+                    None => {
+                        let null = cx.null();
+                        output.set(&mut cx, "numIndexedRows", null)?;
+                    }
+                };
+
+                match unindexed_rows {
+                    Some(x) => {
+                        let i = cx.number(x as f64);
+                        output.set(&mut cx, "numUnindexedRows", i)?;
+                    }
+                    None => {
+                        let null = cx.null();
+                        output.set(&mut cx, "numUnindexedRows", null)?;
+                    }
+                };
+
+                Ok(output)
+            })
+        });
+
+        Ok(promise)
+    }
 }

rust/vectordb/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "vectordb"
-version = "0.3.3"
+version = "0.3.9"
 edition = "2021"
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license = "Apache-2.0"
@@ -21,6 +21,7 @@ object_store = { workspace = true }
 snafu = { workspace = true }
 half = { workspace = true }
 lance = { workspace = true }
+lance-index = { workspace = true }
 lance-linalg = { workspace = true }
 lance-testing = { workspace = true }
 tokio = { version = "1.23", features = ["rt-multi-thread"] }

rust/vectordb/src/database.rs

@@ -161,7 +161,7 @@ impl Database {
     ///
     /// * A [Vec<String>] with all table names.
     pub async fn table_names(&self) -> Result<Vec<String>> {
-        let f = self
+        let mut f = self
             .object_store
             .read_dir(self.base_path.clone())
             .await?
@@ -175,7 +175,8 @@ impl Database {
                 is_lance.unwrap_or(false)
             })
             .filter_map(|p| p.file_stem().and_then(|s| s.to_str().map(String::from)))
-            .collect();
+            .collect::<Vec<String>>();
+        f.sort();
         Ok(f)
     }
 
@@ -312,8 +313,8 @@ mod tests {
         let db = Database::connect(uri).await.unwrap();
         let tables = db.table_names().await.unwrap();
         assert_eq!(tables.len(), 2);
-        assert!(tables.contains(&String::from("table1")));
+        assert!(tables[0].eq(&String::from("table1")));
-        assert!(tables.contains(&String::from("table2")));
+        assert!(tables[1].eq(&String::from("table2")));
     }
 
     #[tokio::test]

rust/vectordb/src/index/vector.rs

@@ -12,9 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use lance::index::vector::ivf::IvfBuildParams;
+use lance::format::{Index, Manifest};
 use lance::index::vector::pq::PQBuildParams;
 use lance::index::vector::VectorIndexParams;
+use lance_index::vector::ivf::IvfBuildParams;
 use lance_linalg::distance::MetricType;
 
 pub trait VectorIndexBuilder {
@@ -98,7 +99,11 @@ impl VectorIndexBuilder for IvfPQIndexBuilder {
         let ivf_params = self.ivf_params.clone().unwrap_or_default();
         let pq_params = self.pq_params.clone().unwrap_or_default();
 
-        VectorIndexParams::with_ivf_pq_params(pq_params.metric_type, ivf_params, pq_params)
+        VectorIndexParams::with_ivf_pq_params(
+            self.metric_type.unwrap_or(MetricType::L2),
+            ivf_params,
+            pq_params,
+        )
     }
 
     fn get_replace(&self) -> bool {
@@ -106,13 +111,34 @@ impl VectorIndexBuilder for IvfPQIndexBuilder {
     }
 }
 
+pub struct VectorIndex {
+    pub columns: Vec<String>,
+    pub index_name: String,
+    pub index_uuid: String,
+}
+
+impl VectorIndex {
+    pub fn new_from_format(manifest: &Manifest, index: &Index) -> VectorIndex {
+        let fields = index
+            .fields
+            .iter()
+            .map(|i| manifest.schema.fields[*i as usize].name.clone())
+            .collect();
+        VectorIndex {
+            columns: fields,
+            index_name: index.name.clone(),
+            index_uuid: index.uuid.to_string(),
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
 
-    use lance::index::vector::ivf::IvfBuildParams;
-    use lance::index::vector::pq::PQBuildParams;
     use lance::index::vector::StageParams;
+    use lance_index::vector::ivf::IvfBuildParams;
+    use lance_index::vector::pq::PQBuildParams;
 
     use crate::index::vector::{IvfPQIndexBuilder, VectorIndexBuilder};
 
@@ -158,7 +184,6 @@ mod tests {
         pq_params.max_iters = 1;
         pq_params.num_bits = 8;
         pq_params.num_sub_vectors = 50;
-        pq_params.metric_type = MetricType::Cosine;
         pq_params.max_opq_iters = 2;
         index_builder.ivf_params(ivf_params);
         index_builder.pq_params(pq_params);
@@ -176,7 +201,6 @@ mod tests {
             assert_eq!(pq_params.max_iters, 1);
             assert_eq!(pq_params.num_bits, 8);
             assert_eq!(pq_params.num_sub_vectors, 50);
-            assert_eq!(pq_params.metric_type, MetricType::Cosine);
             assert_eq!(pq_params.max_opq_iters, 2);
         } else {
             assert!(false, "Expected second stage to be pq")

rust/vectordb/src/io/object_store.rs

@@ -25,7 +25,8 @@ use bytes::Bytes;
 use futures::{stream::BoxStream, FutureExt, StreamExt};
 use lance::io::object_store::WrappingObjectStore;
 use object_store::{
-    path::Path, GetOptions, GetResult, ListResult, MultipartId, ObjectMeta, ObjectStore, Result,
+    path::Path, Error, GetOptions, GetResult, ListResult, MultipartId, ObjectMeta, ObjectStore,
+    Result,
 };
 
 use async_trait::async_trait;
@@ -120,7 +121,10 @@ impl ObjectStore for MirroringObjectStore {
 
     async fn delete(&self, location: &Path) -> Result<()> {
         if !location.primary_only() {
-            self.secondary.delete(location).await?;
+            match self.secondary.delete(location).await {
+                Err(Error::NotFound { .. }) | Ok(_) => {}
+                Err(e) => return Err(e),
+            }
         }
         self.primary.delete(location).await
     }

rust/vectordb/src/query.rs

@@ -32,6 +32,7 @@ pub struct Query {
     pub refine_factor: Option<u32>,
     pub metric_type: Option<MetricType>,
     pub use_index: bool,
+    pub prefilter: bool,
 }
 
 impl Query {
@@ -56,6 +57,7 @@ impl Query {
             use_index: true,
             filter: None,
             select: None,
+            prefilter: false,
         }
     }
 
@@ -74,6 +76,8 @@ impl Query {
         )?;
         scanner.nprobs(self.nprobes);
         scanner.use_index(self.use_index);
+        scanner.prefilter(self.prefilter);
+
         self.select.as_ref().map(|p| scanner.project(p.as_slice()));
         self.filter.as_ref().map(|f| scanner.filter(f));
         self.refine_factor.map(|rf| scanner.refine(rf));
@@ -158,6 +162,11 @@ impl Query {
         self.select = columns;
         self
     }
+
+    pub fn prefilter(mut self, prefilter: bool) -> Query {
+        self.prefilter = prefilter;
+        self
+    }
 }
 
 #[cfg(test)]
@@ -167,7 +176,9 @@ mod tests {
     use super::*;
     use arrow_array::{Float32Array, RecordBatch, RecordBatchIterator, RecordBatchReader};
     use arrow_schema::{DataType, Field as ArrowField, Schema as ArrowSchema};
+    use futures::StreamExt;
     use lance::dataset::Dataset;
+    use lance_testing::datagen::{BatchGenerator, IncrementingInt32, RandomVector};
 
     use crate::query::Query;
 
@@ -200,13 +211,43 @@ mod tests {
 
     #[tokio::test]
     async fn test_execute() {
-        let batches = make_test_batches();
+        let batches = make_non_empty_batches();
-        let ds = Dataset::write(batches, "memory://foo", None).await.unwrap();
+        let ds = Arc::new(Dataset::write(batches, "memory://foo", None).await.unwrap());
 
-        let vector = Float32Array::from_iter_values([0.1; 128]);
+        let vector = Float32Array::from_iter_values([0.1; 4]);
-        let query = Query::new(Arc::new(ds), vector.clone());
+
-        let result = query.execute().await;
+        let query = Query::new(ds.clone(), vector.clone());
-        assert_eq!(result.is_ok(), true);
+        let result = query
+            .limit(10)
+            .filter(Some("id % 2 == 0".to_string()))
+            .execute()
+            .await;
+        let mut stream = result.expect("should have result");
+        // should only have one batch
+        while let Some(batch) = stream.next().await {
+            // post filter should have removed some rows
+            assert!(batch.expect("should be Ok").num_rows() < 10);
+        }
+
+        let query = Query::new(ds, vector.clone());
+        let result = query
+            .limit(10)
+            .filter(Some("id % 2 == 0".to_string()))
+            .prefilter(true)
+            .execute()
+            .await;
+        let mut stream = result.expect("should have result");
+        // should only have one batch
+        while let Some(batch) = stream.next().await {
+            // pre filter should return 10 rows
+            assert!(batch.expect("should be Ok").num_rows() == 10);
+        }
+    }
+
+    fn make_non_empty_batches() -> impl RecordBatchReader + Send + 'static {
+        let vec = Box::new(RandomVector::new().named("vector".to_string()));
+        let id = Box::new(IncrementingInt32::new().named("id".to_string()));
+        BatchGenerator::new().col(vec).col(id).batch(512)
     }
 
     fn make_test_batches() -> impl RecordBatchReader + Send + 'static {

rust/vectordb/src/table.rs

@@ -13,19 +13,23 @@
 // limitations under the License.
 
 use chrono::Duration;
+use lance::dataset::builder::DatasetBuilder;
+use lance_index::IndexType;
 use std::sync::Arc;
 
 use arrow_array::{Float32Array, RecordBatchReader};
 use arrow_schema::SchemaRef;
 use lance::dataset::cleanup::RemovalStats;
-use lance::dataset::optimize::{compact_files, CompactionMetrics, CompactionOptions};
+use lance::dataset::optimize::{
+    compact_files, CompactionMetrics, CompactionOptions, IndexRemapperOptions,
+};
 use lance::dataset::{Dataset, WriteParams};
-use lance::index::IndexType;
+use lance::index::DatasetIndexExt;
 use lance::io::object_store::WrappingObjectStore;
 use std::path::Path;
 
 use crate::error::{Error, Result};
-use crate::index::vector::VectorIndexBuilder;
+use crate::index::vector::{VectorIndex, VectorIndexBuilder};
 use crate::query::Query;
 use crate::utils::{PatchReadParam, PatchWriteParam};
 use crate::WriteMode;
@@ -94,7 +98,10 @@ impl Table {
             Some(wrapper) => params.patch_with_store_wrapper(wrapper)?,
             None => params,
         };
-        let dataset = Dataset::open_with_params(uri, &params)
+
+        let dataset = DatasetBuilder::from_uri(uri)
+            .with_read_params(params)
+            .load()
             .await
             .map_err(|e| match e {
                 lance::Error::DatasetNotFound { .. } => Error::TableNotFound {
@@ -238,8 +245,6 @@ impl Table {
 
     /// Create index on the table.
     pub async fn create_index(&mut self, index_builder: &impl VectorIndexBuilder) -> Result<()> {
-        use lance::index::DatasetIndexExt;
-
         let mut dataset = self.dataset.as_ref().clone();
         dataset
             .create_index(
@@ -257,6 +262,14 @@ impl Table {
         Ok(())
     }
 
+    pub async fn optimize_indices(&mut self) -> Result<()> {
+        let mut dataset = self.dataset.as_ref().clone();
+
+        dataset.optimize_indices().await?;
+
+        Ok(())
+    }
+
     /// Insert records into this Table
     ///
     /// # Arguments
@@ -353,12 +366,45 @@ impl Table {
     /// for faster reads.
     ///
     /// This calls into [lance::dataset::optimize::compact_files].
-    pub async fn compact_files(&mut self, options: CompactionOptions) -> Result<CompactionMetrics> {
+    pub async fn compact_files(
+        &mut self,
+        options: CompactionOptions,
+        remap_options: Option<Arc<dyn IndexRemapperOptions>>,
+    ) -> Result<CompactionMetrics> {
         let mut dataset = self.dataset.as_ref().clone();
-        let metrics = compact_files(&mut dataset, options, None).await?;
+        let metrics = compact_files(&mut dataset, options, remap_options).await?;
         self.dataset = Arc::new(dataset);
         Ok(metrics)
     }
+
+    pub fn count_fragments(&self) -> usize {
+        self.dataset.count_fragments()
+    }
+
+    pub async fn count_deleted_rows(&self) -> Result<usize> {
+        Ok(self.dataset.count_deleted_rows().await?)
+    }
+
+    pub async fn num_small_files(&self, max_rows_per_group: usize) -> usize {
+        self.dataset.num_small_files(max_rows_per_group).await
+    }
+
+    pub async fn count_indexed_rows(&self, index_uuid: &str) -> Result<Option<usize>> {
+        Ok(self.dataset.count_indexed_rows(index_uuid).await?)
+    }
+
+    pub async fn count_unindexed_rows(&self, index_uuid: &str) -> Result<Option<usize>> {
+        Ok(self.dataset.count_unindexed_rows(index_uuid).await?)
+    }
+
+    pub async fn load_indices(&self) -> Result<Vec<VectorIndex>> {
+        let (indices, mf) =
+            futures::try_join!(self.dataset.load_indices(), self.dataset.latest_manifest())?;
+        Ok(indices
+            .iter()
+            .map(|i| VectorIndex::new_from_format(&mf, i))
+            .collect())
+    }
 }
 
 #[cfg(test)]
@@ -373,9 +419,9 @@ mod tests {
     use arrow_data::ArrayDataBuilder;
     use arrow_schema::{DataType, Field, Schema};
     use lance::dataset::{Dataset, WriteMode};
-    use lance::index::vector::ivf::IvfBuildParams;
     use lance::index::vector::pq::PQBuildParams;
     use lance::io::object_store::{ObjectStoreParams, WrappingObjectStore};
+    use lance_index::vector::ivf::IvfBuildParams;
     use rand::Rng;
     use tempfile::tempdir;