[python] Bump version: 0.5.5 → 0.5.6

I think this should work. Need to deploy it to be sure as it can be
tested locally. Can be tested here.

2 things about this solution:
* All pages have a same meta tag, i.e, lancedb banner
* If needed, we can automatically use the first image of each page and
generate meta tags using the ultralytics mkdocs plugin that we did for
this purpose - https://github.com/ultralytics/mkdocs

.cargo/config.toml

@@ -33,3 +33,8 @@ rustflags = ["-C", "target-cpu=haswell", "-C", "target-feature=+avx2,+fma,+f16c"
 
 [target.aarch64-apple-darwin]
 rustflags = ["-C", "target-cpu=apple-m1", "-C", "target-feature=+neon,+fp16,+fhm,+dotprod"]
+
+# Not all Windows systems have the C runtime installed, so this avoids library
+# not found errors on systems that are missing it.
+[target.x86_64-pc-windows-msvc]
+rustflags = ["-Ctarget-feature=+crt-static"]

.github/workflows/npm-publish.yml

@@ -80,10 +80,25 @@ jobs:
           - arch: x86_64
             runner: ubuntu-latest
           - arch: aarch64
-            runner: buildjet-8vcpu-ubuntu-2204-arm
+            # For successful fat LTO builds, we need a large runner to avoid OOM errors.
+            runner: buildjet-16vcpu-ubuntu-2204-arm
     steps:
       - name: Checkout
         uses: actions/checkout@v4
+      # Buildjet aarch64 runners have only 1.5 GB RAM per core, vs 3.5 GB per core for
+      # x86_64 runners. To avoid OOM errors on ARM, we create a swap file.
+      - name: Configure aarch64 build
+        if: ${{ matrix.config.arch == 'aarch64' }}
+        run: |
+          free -h
+          sudo fallocate -l 16G /swapfile
+          sudo chmod 600 /swapfile
+          sudo mkswap /swapfile
+          sudo swapon /swapfile
+          echo "/swapfile swap swap defaults 0 0" >> sudo /etc/fstab
+          # print info
+          swapon --show
+          free -h
       - name: Build Linux Artifacts
         run: |
           bash ci/build_linux_artifacts.sh ${{ matrix.config.arch }}

Cargo.toml

@@ -14,10 +14,10 @@ keywords = ["lancedb", "lance", "database", "vector", "search"]
 categories = ["database-implementations"]
 
 [workspace.dependencies]
-lance = { "version" = "=0.9.15", "features" = ["dynamodb"] }
-lance-index = { "version" = "=0.9.15" }
-lance-linalg = { "version" = "=0.9.15" }
-lance-testing = { "version" = "=0.9.15" }
+lance = { "version" = "=0.9.18", "features" = ["dynamodb"] }
+lance-index = { "version" = "=0.9.18" }
+lance-linalg = { "version" = "=0.9.18" }
+lance-testing = { "version" = "=0.9.18" }
 # Note that this one does not include pyarrow
 arrow = { version = "50.0", optional = false }
 arrow-array = "50.0"

ci/build_linux_artifacts.sh

@@ -13,7 +13,9 @@ docker build \
     .
 popd
 
+# We turn on memory swap to avoid OOM killer
 docker run \
     -v $(pwd):/io -w /io \
+    --memory-swap=-1 \
     lancedb-node-manylinux \
     bash ci/manylinux_node/build.sh $ARCH

docs/mkdocs.yml

@@ -92,16 +92,16 @@ nav:
     - Full-text search: fts.md
     - Hybrid search:
       - Overview: hybrid_search/hybrid_search.md
+      - Comparing Rerankers: hybrid_search/eval.md
       - Airbnb financial data example: notebooks/hybrid_search.ipynb
     - Filtering: sql.md
     - Versioning & Reproducibility: notebooks/reproducibility.ipynb
     - Configuring Storage: guides/storage.md
   - 🧬 Managing embeddings:
     - Overview: embeddings/index.md
-    - Explicit management: embeddings/embedding_explicit.md
-    - Implicit management: embeddings/embedding_functions.md
-    - Available Functions: embeddings/default_embedding_functions.md
-    - Custom Embedding Functions: embeddings/api.md
+    - Embedding functions: embeddings/embedding_functions.md
+    - Available models: embeddings/default_embedding_functions.md
+    - User-defined embedding functions: embeddings/custom_embedding_function.md
     - "Example: Multi-lingual semantic search": notebooks/multi_lingual_example.ipynb
     - "Example: MultiModal CLIP Embeddings": notebooks/DisappearingEmbeddingFunction.ipynb
   - 🔌 Integrations:
@@ -156,16 +156,16 @@ nav:
     - Full-text search: fts.md
     - Hybrid search:
       - Overview: hybrid_search/hybrid_search.md
+      - Comparing Rerankers: hybrid_search/eval.md
       - Airbnb financial data example: notebooks/hybrid_search.ipynb
     - Filtering: sql.md
     - Versioning & Reproducibility: notebooks/reproducibility.ipynb
     - Configuring Storage: guides/storage.md
 - Managing Embeddings:
   - Overview: embeddings/index.md
-  - Explicit management: embeddings/embedding_explicit.md
-  - Implicit management: embeddings/embedding_functions.md
-  - Available Functions: embeddings/default_embedding_functions.md
-  - Custom Embedding Functions: embeddings/api.md
+  - Embedding functions: embeddings/embedding_functions.md
+  - Available models: embeddings/default_embedding_functions.md
+  - User-defined embedding functions: embeddings/custom_embedding_function.md
   - "Example: Multi-lingual semantic search": notebooks/multi_lingual_example.ipynb
   - "Example: MultiModal CLIP Embeddings": notebooks/DisappearingEmbeddingFunction.ipynb
 - Integrations:
@@ -206,6 +206,7 @@ extra_css:
 
 extra_javascript:
   - "extra_js/init_ask_ai_widget.js"
+  - "extra_js/meta_tag.js"
 
 extra:
   analytics:

docs/src/embeddings/embedding_explicit.md

@@ -1,141 +0,0 @@
-In this workflow, you define your own embedding function and pass it as a callable to LanceDB, invoking it in your code to generate the embeddings. Let's look at some examples.
-
-### Hugging Face
-
-!!! note
-    Currently, the Hugging Face method is only supported in the Python SDK.
-
-=== "Python"
-    The most popular open source option is to use the [sentence-transformers](https://www.sbert.net/) 
-    library, which can be installed via pip.
-
-    ```bash
-    pip install sentence-transformers
-    ```
-
-    The example below shows how to use the `paraphrase-albert-small-v2` model to generate embeddings 
-    for a given document.
-
-    ```python
-    from sentence_transformers import SentenceTransformer
-
-    name="paraphrase-albert-small-v2"
-    model = SentenceTransformer(name)
-
-    # used for both training and querying
-    def embed_func(batch):
-        return [model.encode(sentence) for sentence in batch]
-    ```
-
-### OpenAI
-
-Another popular alternative is to use an external API like OpenAI's [embeddings API](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings).
-
-=== "Python"
-      ```python
-        import openai
-        import os
-
-        # Configuring the environment variable OPENAI_API_KEY
-        if "OPENAI_API_KEY" not in os.environ:
-        # OR set the key here as a variable
-        openai.api_key = "sk-..."
-
-        # verify that the API key is working
-        assert len(openai.Model.list()["data"]) > 0
-
-        def embed_func(c):
-            rs = openai.Embedding.create(input=c, engine="text-embedding-ada-002")
-            return [record["embedding"] for record in rs["data"]]
-      ```
-
-=== "JavaScript"
-      ```javascript
-        const lancedb = require("vectordb");
-
-        // You need to provide an OpenAI API key
-        const apiKey = "sk-..."
-        // The embedding function will create embeddings for the 'text' column
-        const embedding = new lancedb.OpenAIEmbeddingFunction('text', apiKey)
-      ```
-
-## Applying an embedding function to data
-
-=== "Python"
-    Using an embedding function, you can apply it to raw data
-    to generate embeddings for each record.
-
-    Say you have a pandas DataFrame with a `text` column that you want embedded,
-    you can use the `with_embeddings` function to generate embeddings and add them to 
-    an existing table.
-
-    ```python
-     import pandas as pd
-     from lancedb.embeddings import with_embeddings
-
-     df = pd.DataFrame(
-        [
-            {"text": "pepperoni"},
-            {"text": "pineapple"}
-        ]
-    )
-     data = with_embeddings(embed_func, df)
-
-     # The output is used to create / append to a table
-     # db.create_table("my_table", data=data)
-    ```
-
-    If your data is in a different column, you can specify the `column` kwarg to `with_embeddings`.
-
-    By default, LanceDB calls the function with batches of 1000 rows. This can be configured
-    using the `batch_size` parameter to `with_embeddings`.
-
-    LanceDB automatically wraps the function with retry and rate-limit logic to ensure the OpenAI
-    API call is reliable.
-
-=== "JavaScript"
-    Using an embedding function, you can apply it to raw data
-    to generate embeddings for each record.
-
-    Simply pass the embedding function created above and LanceDB will use it to generate
-    embeddings for your data.
-
-    ```javascript
-    const db = await lancedb.connect("data/sample-lancedb");
-    const data = [
-    { text: "pepperoni"},
-    { text: "pineapple"}
-    ]
-
-    const table = await db.createTable("vectors", data, embedding)
-    ```
-
-## Querying using an embedding function
-
-!!! warning
-    At query time, you **must** use the same embedding function you used to vectorize your data.
-    If you use a different embedding function, the embeddings will not reside in the same vector
-    space and the results will be nonsensical.
-
-=== "Python"
-     ```python
-     query = "What's the best pizza topping?"
-     query_vector = embed_func([query])[0]
-     results = (
-        tbl.search(query_vector)
-        .limit(10)
-        .to_pandas()
-     )
-     ```
-
-     The above snippet returns a pandas DataFrame with the 10 closest vectors to the query.
-
-=== "JavaScript"
-     ```javascript
-      const results = await table
-        .search("What's the best pizza topping?")
-        .limit(10)
-        .execute()
-     ```
-
-     The above snippet returns an array of records with the top 10 nearest neighbors to the query.

docs/src/embeddings/embedding_functions.md

@@ -3,61 +3,126 @@ Representing multi-modal data as vector embeddings is becoming a standard practi
 For this purpose, LanceDB introduces an **embedding functions API**, that allow you simply set up once, during the configuration stage of your project. After this, the table remembers it, effectively making the embedding functions *disappear in the background* so you don't have to worry about manually passing callables, and instead, simply focus on the rest of your data engineering pipeline.
 
 !!! warning
-    Using the implicit embeddings management approach means that you can forget about the manually passing around embedding
-    functions in your code, as long as you don't intend to change it at a later time. If your embedding function changes,
-    you'll have to re-configure your table with the new embedding function and regenerate the embeddings.
+    Using the embedding function registry means that you don't have to explicitly generate the embeddings yourself. 
+    However, if your embedding function changes, you'll have to re-configure your table with the new embedding function 
+    and regenerate the embeddings. In the future, we plan to support the ability to change the embedding function via
+    table metadata and have LanceDB automatically take care of regenerating the embeddings.
+
 
 ## 1. Define the embedding function
-We have some pre-defined embedding functions in the global registry, with more coming soon. Here's let's an implementation of CLIP as example.
-```
-registry = EmbeddingFunctionRegistry.get_instance()
-clip = registry.get("open-clip").create()
 
-```
-You can also define your own embedding function by implementing the `EmbeddingFunction` abstract base interface. It subclasses Pydantic Model which can be utilized to write complex schemas simply as we'll see next!
+=== "Python"
+    In the LanceDB python SDK, we define a global embedding function registry with
+    many different embedding models and even more coming soon. 
+    Here's let's an implementation of CLIP as example.
+
+    ```python
+    from lancedb.embeddings import get_registry
+
+    registry = get_registry()
+    clip = registry.get("open-clip").create()
+    ```
+
+    You can also define your own embedding function by implementing the `EmbeddingFunction` 
+    abstract base interface. It subclasses Pydantic Model which can be utilized to write complex schemas simply as we'll see next!
+
+=== "JavaScript""
+    In the TypeScript SDK, the choices are more limited. For now, only the OpenAI
+    embedding function is available.
+
+    ```javascript
+    const lancedb = require("vectordb");
+
+    // You need to provide an OpenAI API key
+    const apiKey = "sk-..."
+    // The embedding function will create embeddings for the 'text' column
+    const embedding = new lancedb.OpenAIEmbeddingFunction('text', apiKey)
+    ```
 
 ## 2. Define the data model or schema
-The embedding function defined above abstracts away all the details about the models and dimensions required to define the schema. You can simply set a field as **source** or **vector** column. Here's how:
 
-```python
-class Pets(LanceModel):
-    vector: Vector(clip.ndims) = clip.VectorField()
-    image_uri: str = clip.SourceField()
-```
+=== "Python"
+    The embedding function defined above abstracts away all the details about the models and dimensions required to define the schema. You can simply set a field as **source** or **vector** column. Here's how:
 
-`VectorField` tells LanceDB to use the clip embedding function to generate query embeddings for the `vector` column and `SourceField` ensures that when adding data, we automatically use the specified embedding function to encode `image_uri`.
+    ```python
+    class Pets(LanceModel):
+        vector: Vector(clip.ndims) = clip.VectorField()
+        image_uri: str = clip.SourceField()
+    ```
 
-## 3. Create LanceDB table
-Now that we have chosen/defined our embedding function and the schema, we can create the table:
+    `VectorField` tells LanceDB to use the clip embedding function to generate query embeddings for the `vector` column and `SourceField` ensures that when adding data, we automatically use the specified embedding function to encode `image_uri`.
 
-```python
-db = lancedb.connect("~/lancedb")
-table = db.create_table("pets", schema=Pets)
+=== "JavaScript"
 
-```
+    For the TypeScript SDK, a schema can be inferred from input data, or an explicit
+    Arrow schema can be provided.
 
-That's it! We've provided all the information needed to embed the source and query inputs. We can now forget about the model and dimension details and start to build our VectorDB pipeline.
+## 3. Create table and add data
 
-## 4. Ingest lots of data and query your table
-Any new or incoming data can just be added and it'll be vectorized automatically.
+Now that we have chosen/defined our embedding function and the schema, 
+we can create the table and ingest data without needing to explicitly generate
+the embeddings at all:
 
-```python
-table.add([{"image_uri": u} for u in uris])
-```
+=== "Python"
+    ```python
+    db = lancedb.connect("~/lancedb")
+    table = db.create_table("pets", schema=Pets)
 
-Our OpenCLIP query embedding function supports querying via both text and images:
+    table.add([{"image_uri": u} for u in uris])
+    ```
 
-```python
-result = table.search("dog")
-```
+=== "JavaScript"
 
-Let's query an image:
+    ```javascript
+    const db = await lancedb.connect("data/sample-lancedb");
+    const data = [
+    { text: "pepperoni"},
+    { text: "pineapple"}
+    ]
 
-```python
-p = Path("path/to/images/samoyed_100.jpg")
-query_image = Image.open(p)
-table.search(query_image)
-```
+    const table = await db.createTable("vectors", data, embedding)
+    ```
+
+## 4. Querying your table
+Not only can you forget about the embeddings during ingestion, you also don't
+need to worry about it when you query the table:
+
+=== "Python"
+
+    Our OpenCLIP query embedding function supports querying via both text and images:
+
+    ```python
+    results = (
+        table.search("dog")
+        .limit(10)
+        .to_pandas()
+    )
+    ```
+
+    Or we can search using an image:
+
+    ```python
+    p = Path("path/to/images/samoyed_100.jpg")
+    query_image = Image.open(p)
+    results = (
+        table.search(query_image)
+        .limit(10)
+        .to_pandas()
+    )
+    ```
+
+    Both of the above snippet returns a pandas DataFrame with the 10 closest vectors to the query.
+
+=== "JavaScript"
+
+    ```javascript
+    const results = await table
+    .search("What's the best pizza topping?")
+    .limit(10)
+    .execute()
+    ```    
+    
+    The above snippet returns an array of records with the top 10 nearest neighbors to the query.
 
 ---
 
@@ -100,4 +165,5 @@ rs[2].image
 
 ![](../assets/dog_clip_output.png)
 
-Now that you have the basic idea about implicit management via embedding functions, let's dive deeper into a [custom API](./api.md) that you can use to implement your own embedding functions.
+Now that you have the basic idea about LanceDB embedding functions and the embedding function registry,
+let's dive deeper into defining your own [custom functions](./custom_embedding_function.md).

docs/src/embeddings/index.md

@@ -1,8 +1,14 @@
-Due to the nature of vector embeddings, they can be used to represent any kind of data, from text to images to audio. This makes them a very powerful tool for machine learning practitioners. However, there's no one-size-fits-all solution for generating embeddings - there are many different libraries and APIs (both commercial and open source) that can be used to generate embeddings from structured/unstructured data.
+Due to the nature of vector embeddings, they can be used to represent any kind of data, from text to images to audio. 
+This makes them a very powerful tool for machine learning practitioners. 
+However, there's no one-size-fits-all solution for generating embeddings - there are many different libraries and APIs 
+(both commercial and open source) that can be used to generate embeddings from structured/unstructured data.
 
-LanceDB supports 2 methods of vectorizing your raw data into embeddings.
+LanceDB supports 3 methods of working with embeddings.
 
-1. **Explicit**: By manually calling LanceDB's `with_embedding` function to vectorize your data via an `embed_func` of your choice
-2. **Implicit**: Allow LanceDB to embed the data and queries in the background as they come in, by using the table's `EmbeddingRegistry` information
+1. You can manually generate embeddings for the data and queries. This is done outside of LanceDB.
+2. You can use the built-in [embedding functions](./embedding_functions.md) to embed the data and queries in the background.
+3. For python users, you can define your own [custom embedding function](./custom_embedding_function.md)
+   that extends the default embedding functions.
 
-See the [explicit](embedding_explicit.md) and [implicit](embedding_functions.md) embedding sections for more details.
+For python users, there is also a legacy [with_embeddings API](./legacy.md).
+It is retained for compatibility and will be removed in a future version.

docs/src/embeddings/legacy.md

@@ -0,0 +1,99 @@
+The legacy `with_embeddings` API is for Python only and is deprecated.
+
+### Hugging Face
+
+The most popular open source option is to use the [sentence-transformers](https://www.sbert.net/) 
+library, which can be installed via pip.
+
+```bash
+pip install sentence-transformers
+```
+
+The example below shows how to use the `paraphrase-albert-small-v2` model to generate embeddings 
+for a given document.
+
+```python
+from sentence_transformers import SentenceTransformer
+
+name="paraphrase-albert-small-v2"
+model = SentenceTransformer(name)
+
+# used for both training and querying
+def embed_func(batch):
+    return [model.encode(sentence) for sentence in batch]
+```
+
+
+### OpenAI
+
+Another popular alternative is to use an external API like OpenAI's [embeddings API](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings).
+
+```python
+import openai
+import os
+
+# Configuring the environment variable OPENAI_API_KEY
+if "OPENAI_API_KEY" not in os.environ:
+# OR set the key here as a variable
+openai.api_key = "sk-..."
+
+client = openai.OpenAI()
+
+def embed_func(c):    
+    rs = client.embeddings.create(input=c, model="text-embedding-ada-002")
+    return [record.embedding for record in rs["data"]]
+```
+
+
+## Applying an embedding function to data
+
+Using an embedding function, you can apply it to raw data
+to generate embeddings for each record.
+
+Say you have a pandas DataFrame with a `text` column that you want embedded,
+you can use the `with_embeddings` function to generate embeddings and add them to 
+an existing table.
+
+```python
+    import pandas as pd
+    from lancedb.embeddings import with_embeddings
+
+    df = pd.DataFrame(
+        [
+            {"text": "pepperoni"},
+            {"text": "pineapple"}
+        ]
+    )
+    data = with_embeddings(embed_func, df)
+
+    # The output is used to create / append to a table
+    tbl = db.create_table("my_table", data=data)
+```
+
+If your data is in a different column, you can specify the `column` kwarg to `with_embeddings`.
+
+By default, LanceDB calls the function with batches of 1000 rows. This can be configured
+using the `batch_size` parameter to `with_embeddings`.
+
+LanceDB automatically wraps the function with retry and rate-limit logic to ensure the OpenAI
+API call is reliable.
+
+## Querying using an embedding function
+
+!!! warning
+    At query time, you **must** use the same embedding function you used to vectorize your data.
+    If you use a different embedding function, the embeddings will not reside in the same vector
+    space and the results will be nonsensical.
+
+=== "Python"
+     ```python
+     query = "What's the best pizza topping?"
+     query_vector = embed_func([query])[0]
+     results = (
+        tbl.search(query_vector)
+        .limit(10)
+        .to_pandas()
+     )
+     ```
+
+     The above snippet returns a pandas DataFrame with the 10 closest vectors to the query.

docs/src/examples/modal_langchain.py

@@ -1,6 +1,5 @@
 import pickle
 import re
-import sys
 import zipfile
 from pathlib import Path
 

docs/src/extra_js/meta_tag.js

@@ -0,0 +1,6 @@
+window.addEventListener('load', function() {
+    var meta = document.createElement('meta');
+    meta.setAttribute('property', 'og:image');
+    meta.setAttribute('content', '/assets/lancedb_and_lance.png');
+    document.head.appendChild(meta);
+  });

docs/src/hybrid_search/eval.md

@@ -0,0 +1,49 @@
+# Hybrid Search
+
+Hybrid Search is a broad (often misused) term. It can mean anything from combining multiple methods for searching, to applying ranking methods to better sort the results. In this blog, we use the definition of "hybrid search" to mean using a combination of keyword-based and vector search.
+
+## The challenge of (re)ranking search results
+Once you have a group of the most relevant search results from multiple search sources, you'd likely standardize the score and rank them accordingly. This process can also be seen as another independent step - reranking.
+There are two approaches for reranking search results from multiple sources.
+* <b>Score-based</b>: Calculate final relevance scores based on a weighted linear combination of individual search algorithm scores. Example - Weighted linear combination of semantic search & keyword-based search results.
+* <b>Relevance-based</b>: Discards the existing scores and calculates the relevance of each search result - query pair. Example - Cross Encoder models
+
+Even though there are many strategies for reranking search results, none works for all cases. Moreover, evaluating them itself is a challenge. Also, reranking can be dataset, application specific so it's hard to generalize.
+
+### Example evaluation of hybrid search with Reranking
+
+Here's some evaluation numbers from experiment comparing these re-rankers on about 800 queries. It is modified version of an evaluation script from [llama-index](https://github.com/run-llama/finetune-embedding/blob/main/evaluate.ipynb) that measures hit-rate at top-k.
+
+<b> With OpenAI ada2 embedding </b>
+
+Vector Search baseline - `0.64`
+
+| Reranker | Top-3 | Top-5 | Top-10 |
+| --- | --- | --- | --- |
+| Linear Combination | `0.73` | `0.74` | `0.85` |
+| Cross Encoder | `0.71` | `0.70` | `0.77` |
+| Cohere | `0.81` | `0.81` | `0.85` |
+| ColBERT | `0.68` | `0.68` | `0.73` |
+
+<p>
+<img src="https://github.com/AyushExel/assets/assets/15766192/d57b1780-ef27-414c-a5c3-73bee7808a45">
+</p>
+
+<b> With OpenAI embedding-v3-small </b>
+
+Vector Search baseline - `0.59`
+
+| Reranker | Top-3 | Top-5 | Top-10 |
+| --- | --- | --- | --- |
+| Linear Combination | `0.68` | `0.70` | `0.84` |
+| Cross Encoder | `0.72` | `0.72` | `0.79` |
+| Cohere | `0.79` | `0.79` | `0.84` |
+| ColBERT | `0.70` | `0.70` | `0.76` |
+
+<p>
+<img src="https://github.com/AyushExel/assets/assets/15766192/259adfd2-6ec6-4df6-a77d-1456598970dd">
+</p>
+
+### Conclusion
+
+The results show that the reranking methods are able to improve the search results. However, the improvement is not consistent across all rerankers. The choice of reranker depends on the dataset and the application. It is also important to note that the reranking methods are not a replacement for the search methods. They are complementary and should be used together to get the best results. The speed to recall tradeoff is also an important factor to consider when choosing the reranker.

docs/src/notebooks/DisappearingEmbeddingFunction.ipynb

@@ -290,7 +290,7 @@
         "from lancedb.pydantic import LanceModel, Vector\n",
         "\n",
         "class Pets(LanceModel):\n",
-        "    vector: Vector(clip.ndims) = clip.VectorField()\n",
+        "    vector: Vector(clip.ndims()) = clip.VectorField()\n",
         "    image_uri: str = clip.SourceField()\n",
         "\n",
         "    @property\n",
@@ -360,7 +360,7 @@
         "    table = db.create_table(\"pets\", schema=Pets)\n",
         "    # use a sampling of 1000 images\n",
         "    p = Path(\"~/Downloads/images\").expanduser()\n",
-        "    uris = [str(f) for f in p.iterdir()]\n",
+        "    uris = [str(f) for f in p.glob(\"*.jpg\")]\n",
         "    uris = sample(uris, 1000)\n",
         "    table.add(pd.DataFrame({\"image_uri\": uris}))"
       ]
@@ -543,7 +543,7 @@
       ],
       "source": [
         "from PIL import Image\n",
-        "p = Path(\"/Users/changshe/Downloads/images/samoyed_100.jpg\")\n",
+        "p = Path(\"~/Downloads/images/samoyed_100.jpg\").expanduser()\n",
         "query_image = Image.open(p)\n",
         "query_image"
       ]

docs/src/notebooks/diffusiondb/datagen.py

@@ -23,10 +23,8 @@ from multiprocessing import Pool
 import lance
 import pyarrow as pa
 from datasets import load_dataset
-from PIL import Image
 from transformers import CLIPModel, CLIPProcessor, CLIPTokenizerFast
 
-import lancedb
 
 MODEL_ID = "openai/clip-vit-base-patch32"
 

docs/src/python/duckdb.md

@@ -1,6 +1,9 @@
 # DuckDB
 
-LanceDB is very well-integrated with [DuckDB](https://duckdb.org/), an in-process SQL OLAP database. This integration is done via [Arrow](https://duckdb.org/docs/guides/python/sql_on_arrow) .
+In Python, LanceDB tables can also be queried with [DuckDB](https://duckdb.org/), an in-process SQL OLAP database. This means you can write complex SQL queries to analyze your data in LanceDB.
+
+This integration is done via [Apache Arrow](https://duckdb.org/docs/guides/python/sql_on_arrow), which provides zero-copy data sharing between LanceDB and DuckDB. DuckDB is capable of passing down column selections and basic filters to LanceDB, reducing the amount of data that needs to be scanned to perform your query. Finally, the integration allows streaming data from LanceDB tables, allowing you to aggregate tables that won't fit into memory. All of this uses the same mechanism described in DuckDB's blog post *[DuckDB quacks Arrow](https://duckdb.org/2021/12/03/duck-arrow.html)*.
+
 
 We can demonstrate this by first installing `duckdb` and `lancedb`.
 
@@ -19,14 +22,15 @@ data = [
     {"vector": [5.9, 26.5], "item": "bar", "price": 20.0}
 ]
 table = db.create_table("pd_table", data=data)
-arrow_table = table.to_arrow()
 ```
 
-DuckDB can directly query the `pyarrow.Table` object:
+To query the table, first call `to_lance` to convert the table to a "dataset", which is an object that can be queried by DuckDB. Then all you need to do is reference that dataset by the same name in your SQL query.
 
 ```python
 import duckdb
 
+arrow_table = table.to_lance()
+
 duckdb.query("SELECT * FROM arrow_table")
 ```
 

node/package-lock.json

@@ -328,6 +328,66 @@
         "@jridgewell/sourcemap-codec": "^1.4.10"
       }
     },
+    "node_modules/@lancedb/vectordb-darwin-arm64": {
+      "version": "0.4.10",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.4.10.tgz",
+      "integrity": "sha512-y/uHOGb0g15pvqv5tdTyZ6oN+0QVpBmZDzKFWW6pPbuSZjB2uPqcs+ti0RB+AUdmS21kavVQqaNsw/HLKEGrHA==",
+      "cpu": [
+        "arm64"
+      ],
+      "optional": true,
+      "os": [
+        "darwin"
+      ]
+    },
+    "node_modules/@lancedb/vectordb-darwin-x64": {
+      "version": "0.4.10",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.4.10.tgz",
+      "integrity": "sha512-XbfR58OkQpAe0xMSTrwJh9ZjGSzG9EZ7zwO6HfYem8PxcLYAcC6eWRWoSG/T0uObyrPTcYYyvHsp0eNQWYBFAQ==",
+      "cpu": [
+        "x64"
+      ],
+      "optional": true,
+      "os": [
+        "darwin"
+      ]
+    },
+    "node_modules/@lancedb/vectordb-linux-arm64-gnu": {
+      "version": "0.4.10",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.4.10.tgz",
+      "integrity": "sha512-x40WKH9b+KxorRmKr9G7fv8p5mMj8QJQvRMA0v6v+nbZHr2FLlAZV+9mvhHOnm4AGIkPP5335cUgv6Qz6hgwkQ==",
+      "cpu": [
+        "arm64"
+      ],
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@lancedb/vectordb-linux-x64-gnu": {
+      "version": "0.4.10",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.4.10.tgz",
+      "integrity": "sha512-CTGPpuzlqq2nVjUxI9gAJOT1oBANIovtIaFsOmBSnEAHgX7oeAxKy2b6L/kJzsgqSzvR5vfLwYcWFrr6ZmBxSA==",
+      "cpu": [
+        "x64"
+      ],
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@lancedb/vectordb-win32-x64-msvc": {
+      "version": "0.4.10",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.4.10.tgz",
+      "integrity": "sha512-Fd7r74coZyrKzkfXg4WthqOL+uKyJyPTia6imcrMNqKOlTGdKmHf02Qi2QxWZrFaabkRYo4Tpn5FeRJ3yYX8CA==",
+      "cpu": [
+        "x64"
+      ],
+      "optional": true,
+      "os": [
+        "win32"
+      ]
+    },
     "node_modules/@neon-rs/cli": {
       "version": "0.0.160",
       "resolved": "https://registry.npmjs.org/@neon-rs/cli/-/cli-0.0.160.tgz",

nodejs/vectordb/native.d.ts

@@ -73,7 +73,7 @@ export class Table {
   /** Return Schema as empty Arrow IPC file. */
   schema(): Buffer
   add(buf: Buffer): Promise<void>
-  countRows(filter?: string): Promise<bigint>
+  countRows(filter?: string | undefined | null): Promise<bigint>
   delete(predicate: string): Promise<void>
   createIndex(): IndexBuilder
   query(): Query

python/.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.5.5
+current_version = 0.5.6
 commit = True
 message = [python] Bump version: {current_version} → {new_version}
 tag = True

python/lancedb/__init__.py

@@ -13,8 +13,9 @@
 
 import importlib.metadata
 import os
+from concurrent.futures import ThreadPoolExecutor
 from datetime import timedelta
-from typing import Optional
+from typing import Optional, Union
 
 __version__ = importlib.metadata.version("lancedb")
 
@@ -32,6 +33,7 @@ def connect(
     region: str = "us-east-1",
     host_override: Optional[str] = None,
     read_consistency_interval: Optional[timedelta] = None,
+    request_thread_pool: Optional[Union[int, ThreadPoolExecutor]] = None,
 ) -> DBConnection:
     """Connect to a LanceDB database.
 
@@ -58,7 +60,14 @@ def connect(
         the last check, then the table will be checked for updates. Note: this
         consistency only applies to read operations. Write operations are
         always consistent.
-
+    request_thread_pool: int or ThreadPoolExecutor, optional
+        The thread pool to use for making batch requests to the LanceDB Cloud API.
+        If an integer, then a ThreadPoolExecutor will be created with that
+        number of threads. If None, then a ThreadPoolExecutor will be created
+        with the default number of threads. If a ThreadPoolExecutor, then that
+        executor will be used for making requests. This is for LanceDB Cloud
+        only and is only used when making batch requests (i.e., passing in
+        multiple queries to the search method at once).
 
     Examples
     --------
@@ -86,5 +95,9 @@ def connect(
             api_key = os.environ.get("LANCEDB_API_KEY")
         if api_key is None:
             raise ValueError(f"api_key is required to connected LanceDB cloud: {uri}")
-        return RemoteDBConnection(uri, api_key, region, host_override)
+        if isinstance(request_thread_pool, int):
+            request_thread_pool = ThreadPoolExecutor(request_thread_pool)
+        return RemoteDBConnection(
+            uri, api_key, region, host_override, request_thread_pool=request_thread_pool
+        )
     return LanceDBConnection(uri, read_consistency_interval=read_consistency_interval)

python/lancedb/embeddings/utils.py

@@ -26,7 +26,7 @@ import pyarrow as pa
 from lance.vector import vec_to_table
 from retry import retry
 
-from ..util import safe_import_pandas
+from ..util import deprecated, safe_import_pandas
 from ..utils.general import LOGGER
 
 pd = safe_import_pandas()
@@ -38,6 +38,7 @@ IMAGES = Union[
 ]
 
 
+@deprecated
 def with_embeddings(
     func: Callable,
     data: DATA,

python/lancedb/remote/db.py

@@ -14,6 +14,7 @@
 import inspect
 import logging
 import uuid
+from concurrent.futures import ThreadPoolExecutor
 from typing import Iterable, List, Optional, Union
 from urllib.parse import urlparse
 
@@ -39,6 +40,7 @@ class RemoteDBConnection(DBConnection):
         api_key: str,
         region: str,
         host_override: Optional[str] = None,
+        request_thread_pool: Optional[ThreadPoolExecutor] = None,
     ):
         """Connect to a remote LanceDB database."""
         parsed = urlparse(db_url)
@@ -49,6 +51,7 @@ class RemoteDBConnection(DBConnection):
         self._client = RestfulLanceDBClient(
             self.db_name, region, api_key, host_override
         )
+        self._request_thread_pool = request_thread_pool
 
     def __repr__(self) -> str:
         return f"RemoteConnect(name={self.db_name})"

python/lancedb/remote/table.py

@@ -13,6 +13,7 @@
 
 import logging
 import uuid
+from concurrent.futures import Future
 from functools import cached_property
 from typing import Dict, Optional, Union
 
@@ -270,15 +271,28 @@ class RemoteTable(Table):
             and len(query.vector) > 0
             and not isinstance(query.vector[0], float)
         ):
+            if self._conn._request_thread_pool is None:
+
+                def submit(name, q):
+                    f = Future()
+                    f.set_result(self._conn._client.query(name, q))
+                    return f
+            else:
+
+                def submit(name, q):
+                    return self._conn._request_thread_pool.submit(
+                        self._conn._client.query, name, q
+                    )
+
             results = []
             for v in query.vector:
                 v = list(v)
                 q = query.copy()
                 q.vector = v
-                results.append(self._conn._client.query(self._name, q))
+                results.append(submit(self._name, q))
 
             return pa.concat_tables(
-                [add_index(r.to_arrow(), i) for i, r in enumerate(results)]
+                [add_index(r.result().to_arrow(), i) for i, r in enumerate(results)]
             )
         else:
             result = self._conn._client.query(self._name, query)

python/lancedb/util.py

@@ -11,9 +11,11 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 
+import functools
 import importlib
 import os
 import pathlib
+import warnings
 from datetime import date, datetime
 from functools import singledispatch
 from typing import Tuple, Union
@@ -239,3 +241,25 @@ def _(value: list):
 @value_to_sql.register(np.ndarray)
 def _(value: np.ndarray):
     return value_to_sql(value.tolist())
+
+
+def deprecated(func):
+    """This is a decorator which can be used to mark functions
+    as deprecated. It will result in a warning being emitted
+    when the function is used."""
+
+    @functools.wraps(func)
+    def new_func(*args, **kwargs):
+        warnings.simplefilter("always", DeprecationWarning)  # turn off filter
+        warnings.warn(
+            (
+                f"Function {func.__name__} is deprecated and will be "
+                "removed in a future version"
+            ),
+            category=DeprecationWarning,
+            stacklevel=2,
+        )
+        warnings.simplefilter("default", DeprecationWarning)  # reset filter
+        return func(*args, **kwargs)
+
+    return new_func

python/pyproject.toml

@@ -1,9 +1,9 @@
 [project]
 name = "lancedb"
-version = "0.5.5"
+version = "0.5.6"
 dependencies = [
     "deprecation",
-    "pylance==0.9.15",
+    "pylance==0.9.16",
     "ratelimiter~=1.0",
     "retry>=0.9.2",
     "tqdm>=4.27.0",

python/tests/test_table.py

@@ -803,10 +803,8 @@ def test_count_rows(db):
     assert table.count_rows(filter="text='bar'") == 1
 
 
-def test_hybrid_search(db):
-    # hardcoding temporarily.. this test is failing with tmp_path mockdb.
-    # Probably not being parsed right by the fts
-    db = MockDB("~/lancedb_")
+def test_hybrid_search(db, tmp_path):
+    db = MockDB(str(tmp_path))
     # Create a LanceDB table schema with a vector and a text column
     emb = EmbeddingFunctionRegistry.get_instance().get("test")()