Bump version: 0.4.12 → 0.4.13

This PR fixes https://github.com/lancedb/lancedb/issues/1112. It turned
out that K-means is currently used internally, so I figured adding that
context to the docs would be nice.

.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.4.12
+current_version = 0.4.13
 commit = True
 message = Bump version: {current_version} → {new_version}
 tag = True

Cargo.toml

@@ -14,10 +14,10 @@ keywords = ["lancedb", "lance", "database", "vector", "search"]
 categories = ["database-implementations"]
 
 [workspace.dependencies]
-lance = { "version" = "=0.10.2", "features" = ["dynamodb"] }
-lance-index = { "version" = "=0.10.2" }
-lance-linalg = { "version" = "=0.10.2" }
-lance-testing = { "version" = "=0.10.2" }
+lance = { "version" = "=0.10.4", "features" = ["dynamodb"] }
+lance-index = { "version" = "=0.10.4" }
+lance-linalg = { "version" = "=0.10.4" }
+lance-testing = { "version" = "=0.10.4" }
 # Note that this one does not include pyarrow
 arrow = { version = "50.0", optional = false }
 arrow-array = "50.0"

docs/src/concepts/index_ivfpq.md

@@ -31,7 +31,7 @@ As an example, consider starting with 128-dimensional vector consisting of 32-bi
 
 While PQ helps with reducing the size of the index, IVF primarily addresses search performance. The primary purpose of an inverted file index is to facilitate rapid and effective nearest neighbor search by narrowing down the search space.
 
-In IVF, the PQ vector space is divided into *Voronoi cells*, which are essentially partitions that consist of all the points in the space that are within a threshold distance of the given region's seed point. These seed points are used to create an inverted index that correlates each centroid with a list of vectors in the space, allowing a search to be restricted to just a subset of vectors in the index.
+In IVF, the PQ vector space is divided into *Voronoi cells*, which are essentially partitions that consist of all the points in the space that are within a threshold distance of the given region's seed point. These seed points are initialized by running K-means over the stored vectors. The centroids of K-means turn into the seed points which then each define a region. These regions are then are used to create an inverted index that correlates each centroid with a list of vectors in the space, allowing a search to be restricted to just a subset of vectors in the index.
 
 ![](../assets/ivfpq_ivf_desc.webp)
 

docs/src/embeddings/default_embedding_functions.md

@@ -224,7 +224,6 @@ This embedding function supports ingesting images as both bytes and urls. You ca
 !!! info
     LanceDB supports ingesting images directly from accessible links.
 
-
 ```python
 
 db = lancedb.connect(tmp_path)
@@ -290,4 +289,67 @@ print(actual.label)
 
 ```
 
+### Imagebind embeddings
+We have support for [imagebind](https://github.com/facebookresearch/ImageBind) model embeddings. You can download our version of the packaged model via - `pip install imagebind-packaged==0.1.2`.
+
+This function is registered as `imagebind` and supports Audio, Video and Text modalities(extending to Thermal,Depth,IMU data):
+
+| Parameter | Type | Default Value | Description |
+|---|---|---|---|
+| `name` | `str` | `"imagebind_huge"` | Name of the model. |
+| `device` | `str` | `"cpu"` | The device to run the model on. Can be `"cpu"` or `"gpu"`. |
+| `normalize` | `bool` | `False` | set to `True` to normalize your inputs before model ingestion. |
+
+Below is an example demonstrating how the API works:
+
+```python
+db = lancedb.connect(tmp_path)
+registry = EmbeddingFunctionRegistry.get_instance()
+func = registry.get("imagebind").create()
+
+class ImageBindModel(LanceModel):
+    text: str
+    image_uri: str = func.SourceField()
+    audio_path: str
+    vector: Vector(func.ndims()) = func.VectorField()
+
+# add locally accessible image paths
+text_list=["A dog.", "A car", "A bird"]
+image_paths=[".assets/dog_image.jpg", ".assets/car_image.jpg", ".assets/bird_image.jpg"]
+audio_paths=[".assets/dog_audio.wav", ".assets/car_audio.wav", ".assets/bird_audio.wav"]
+
+# Load data
+inputs = [
+    {"text": a, "audio_path": b, "image_uri": c}
+    for a, b, c in zip(text_list, audio_paths, image_paths)
+]
+
+#create table and add data
+table = db.create_table("img_bind", schema=ImageBindModel)
+table.add(inputs)
+```
+
+Now, we can search using any modality:
+
+#### image search
+```python
+query_image = "./assets/dog_image2.jpg" #download an image and enter that path here
+actual = table.search(query_image).limit(1).to_pydantic(ImageBindModel)[0]
+print(actual.text == "dog")
+```
+#### audio search
+
+```python
+query_audio = "./assets/car_audio2.wav" #download an audio clip and enter path here
+actual = table.search(query_audio).limit(1).to_pydantic(ImageBindModel)[0]
+print(actual.text == "car")
+```
+#### Text search
+You can add any input query and fetch the result as follows:
+```python
+query = "an animal which flies and tweets" 
+actual = table.search(query).limit(1).to_pydantic(ImageBindModel)[0]
+print(actual.text == "bird")
+```
+
 If you have any questions about the embeddings API, supported models, or see a relevant model missing, please raise an issue [on GitHub](https://github.com/lancedb/lancedb/issues).

node/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vectordb",
-  "version": "0.4.12",
+  "version": "0.4.13",
   "description": " Serverless, low-latency vector database for AI applications",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
@@ -88,10 +88,10 @@
     }
   },
   "optionalDependencies": {
-    "@lancedb/vectordb-darwin-arm64": "0.4.12",
-    "@lancedb/vectordb-darwin-x64": "0.4.12",
-    "@lancedb/vectordb-linux-arm64-gnu": "0.4.12",
-    "@lancedb/vectordb-linux-x64-gnu": "0.4.12",
-    "@lancedb/vectordb-win32-x64-msvc": "0.4.12"
+    "@lancedb/vectordb-darwin-arm64": "0.4.13",
+    "@lancedb/vectordb-darwin-x64": "0.4.13",
+    "@lancedb/vectordb-linux-arm64-gnu": "0.4.13",
+    "@lancedb/vectordb-linux-x64-gnu": "0.4.13",
+    "@lancedb/vectordb-win32-x64-msvc": "0.4.13"
   }
 }

python/pyproject.toml

@@ -3,7 +3,7 @@ name = "lancedb"
 version = "0.6.3"
 dependencies = [
     "deprecation",
-    "pylance==0.10.2",
+    "pylance==0.10.4",
     "ratelimiter~=1.0",
     "retry>=0.9.2",
     "tqdm>=4.27.0",
@@ -81,6 +81,7 @@ embeddings = [
     "awscli>=1.29.57",
     "botocore>=1.31.57",
 ]
+azure = ["adlfs>=2024.2.0"]
 
 [tool.maturin]
 python-source = "python"

python/python/lancedb/embeddings/imagebind.py

@@ -31,7 +31,7 @@ class ImageBindEmbeddings(EmbeddingFunction):
     six different modalities: images, text, audio, depth, thermal, and IMU data
 
     to download package, run :
-        `pip install imagebind@git+https://github.com/raghavdixit99/ImageBind`
+        `pip install imagebind-packaged==0.1.2`
     """
 
     name: str = "imagebind_huge"

python/python/lancedb/embeddings/openai.py

@@ -113,5 +113,5 @@ class OpenAIEmbeddings(TextEmbeddingFunction):
         if self.organization:
             kwargs["organization"] = self.organization
         if self.api_key:
-            kwargs["api_key"] = self
+            kwargs["api_key"] = self.api_key
         return openai.OpenAI(**kwargs)

python/python/lancedb/remote/table.py

@@ -68,10 +68,16 @@ class RemoteTable(Table):
 
     def list_indices(self):
         """List all the indices on the table"""
-        print(self._name)
         resp = self._conn._client.post(f"/v1/table/{self._name}/index/list/")
         return resp
 
+    def index_stats(self, index_uuid: str):
+        """List all the indices on the table"""
+        resp = self._conn._client.post(
+            f"/v1/table/{self._name}/index/{index_uuid}/stats/"
+        )
+        return resp
+
     def create_scalar_index(
         self,
         column: str,

python/python/lancedb/table.py

@@ -118,7 +118,8 @@ def _append_vector_col(data: pa.Table, metadata: dict, schema: Optional[pa.Schem
     functions = EmbeddingFunctionRegistry.get_instance().parse_functions(metadata)
     for vector_column, conf in functions.items():
         func = conf.function
-        if vector_column not in data.column_names:
+        no_vector_column = vector_column not in data.column_names
+        if no_vector_column or pc.all(pc.is_null(data[vector_column])).as_py():
             col_data = func.compute_source_embeddings_with_retry(
                 data[conf.source_column]
             )
@@ -126,9 +127,16 @@ def _append_vector_col(data: pa.Table, metadata: dict, schema: Optional[pa.Schem
                 dtype = schema.field(vector_column).type
             else:
                 dtype = pa.list_(pa.float32(), len(col_data[0]))
-            data = data.append_column(
-                pa.field(vector_column, type=dtype), pa.array(col_data, type=dtype)
-            )
+            if no_vector_column:
+                data = data.append_column(
+                    pa.field(vector_column, type=dtype), pa.array(col_data, type=dtype)
+                )
+            else:
+                data = data.set_column(
+                    data.column_names.index(vector_column),
+                    pa.field(vector_column, type=dtype),
+                    pa.array(col_data, type=dtype),
+                )
     return data
 
 

python/python/lancedb/util.py

@@ -26,6 +26,18 @@ import pyarrow as pa
 import pyarrow.fs as pa_fs
 
 
+def safe_import_adlfs():
+    try:
+        import adlfs
+
+        return adlfs
+    except ImportError:
+        return None
+
+
+adlfs = safe_import_adlfs()
+
+
 def get_uri_scheme(uri: str) -> str:
     """
     Get the scheme of a URI. If the URI does not have a scheme, assume it is a file URI.
@@ -92,6 +104,17 @@ def fs_from_uri(uri: str) -> Tuple[pa_fs.FileSystem, str]:
         path = get_uri_location(uri)
         return fs, path
 
+    elif get_uri_scheme(uri) == "az" and adlfs is not None:
+        az_blob_fs = adlfs.AzureBlobFileSystem(
+            account_name=os.environ.get("AZURE_STORAGE_ACCOUNT_NAME"),
+            account_key=os.environ.get("AZURE_STORAGE_ACCOUNT_KEY"),
+        )
+
+        fs = pa_fs.PyFileSystem(pa_fs.FSSpecHandler(az_blob_fs))
+
+        path = get_uri_location(uri)
+        return fs, path
+
     return pa_fs.FileSystem.from_uri(uri)
 
 

python/python/lancedb/utils/events.py

@@ -69,7 +69,7 @@ class _Events:
         self.throttled_event_names = ["search_table"]
         self.throttled_events = set()
         self.max_events = 5  # max events to store in memory
-        self.rate_limit = 60.0 * 5  # rate limit (seconds)
+        self.rate_limit = 60.0 * 60.0  # rate limit (seconds)
         self.time = 0.0
 
         if is_git_dir():

python/python/tests/test_embeddings.py

@@ -11,6 +11,7 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 import sys
+from typing import List, Union
 
 import lance
 import lancedb
@@ -23,6 +24,8 @@ from lancedb.embeddings import (
     EmbeddingFunctionRegistry,
     with_embeddings,
 )
+from lancedb.embeddings.base import TextEmbeddingFunction
+from lancedb.embeddings.registry import get_registry, register
 from lancedb.pydantic import LanceModel, Vector
 
 
@@ -112,3 +115,34 @@ def test_embedding_function_rate_limit(tmp_path):
     table.add([{"text": "hello world"}])
     table.add([{"text": "hello world"}])
     assert len(table) == 2
+
+
+def test_add_optional_vector(tmp_path):
+    @register("mock-embedding")
+    class MockEmbeddingFunction(TextEmbeddingFunction):
+        def ndims(self):
+            return 128
+
+        def generate_embeddings(
+            self, texts: Union[List[str], np.ndarray]
+        ) -> List[np.array]:
+            """
+            Generate the embeddings for the given texts
+            """
+            return [np.random.randn(self.ndims()).tolist() for _ in range(len(texts))]
+
+    registry = get_registry()
+    model = registry.get("mock-embedding").create()
+
+    class LanceSchema(LanceModel):
+        id: str
+        vector: Vector(model.ndims()) = model.VectorField(default=None)
+        text: str = model.SourceField()
+
+    db = lancedb.connect(tmp_path)
+    tbl = db.create_table("optional_vector", schema=LanceSchema)
+
+    # add works
+    expected = LanceSchema(id="id", text="text")
+    tbl.add([expected])
+    assert not (np.abs(tbl.to_pandas()["vector"][0]) < 1e-6).all()

python/python/tests/test_io.py

@@ -16,16 +16,35 @@ import os
 import lancedb
 import pytest
 
+# AWS:
 # You need to setup AWS credentials an a base path to run this test. Example
 #    AWS_PROFILE=default TEST_S3_BASE_URL=s3://my_bucket/dataset pytest tests/test_io.py
+#
+# Azure:
+# You need to setup Azure credentials an a base path to run this test. Example
+#   export AZURE_STORAGE_ACCOUNT_NAME="<account>"
+#   export AZURE_STORAGE_ACCOUNT_KEY="<key>"
+#   export REMOTE_BASE_URL=az://my_blob/dataset
+#   pytest tests/test_io.py
+
+
+@pytest.fixture(autouse=True, scope="module")
+def setup():
+    yield
+
+    if remote_url := os.environ.get("REMOTE_BASE_URL"):
+        db = lancedb.connect(remote_url)
+
+        for table in db.table_names():
+            db.drop_table(table)
 
 
 @pytest.mark.skipif(
-    (os.environ.get("TEST_S3_BASE_URL") is None),
-    reason="please setup s3 base url",
+    (os.environ.get("REMOTE_BASE_URL") is None),
+    reason="please setup remote base url",
 )
-def test_s3_io():
-    db = lancedb.connect(os.environ.get("TEST_S3_BASE_URL"))
+def test_remote_io():
+    db = lancedb.connect(os.environ.get("REMOTE_BASE_URL"))
     assert db.table_names() == []
 
     table = db.create_table(

rust/ffi/node/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-node"
-version = "0.4.12"
+version = "0.4.13"
 description = "Serverless, low-latency vector database for AI applications"
 license.workspace = true
 edition.workspace = true

rust/lancedb/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.4.12"
+version = "0.4.13"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true