SVS 0.5.0

CMakeLists.txt

@@ -23,7 +23,7 @@ project(svs
     # - /bindings/python/tests/test_common.py
     # Manually keep in-sync with:
     # - /bindings/python/setup.py
-    VERSION 0.4.0
+    VERSION 0.5.0
 )
 
 set(SVS_LIB svs_devel)

HISTORY.md

@@ -1,3 +1,25 @@
+# SVS 0.4.0 Release Notes
+
+## Additions and Changes
+
+* Native serialization support for Indexes (#300)
+
+* Clang-21 compiler support with gcc-11 headers for shared library builds
+
+* Amazonlinux2 support for shared/static library dropped
+
+* Consolidated to publish only minimal necessary shared library (14 -> 6) and runtime bindings (2 -> 1) release artifacts
+
+* Adaptive batch size heuristic for filtered search in Vamana, with new `filter_stop` early-exit parameter (#309)
+
+* IVF runtime support for `IDFilter`-based filtered search and dynamic `set_intra_query_threads` adjustment after `train()`/`load()` (#322)
+
+* IVF inter-query thread pool initialization now validates and clamps thread count against the number of centroids (#329)
+
+* In-memory (mmap) stream index mapping with new `map_to_file` and `map_to_memory` APIs on `FlatIndex` and `VamanaIndex` (#310)
+
+* Add get_distance and reconstruct_at to VamanaIndex API (#315)
+
 # SVS 0.3.0 Release Notes
 
 ## Additions and Changes

NEWS.md

@@ -1,21 +1,29 @@
-# SVS 0.4.0 Release Notes
+# SVS 0.5.0 Release Notes
 
 ## Additions and Changes
 
-* Native serialization support for Indexes (#300)
+* New C API for SVS — index construction, search, and lifecycle management from C, including filtered TopK search, LeanVec OOD training, memory accounting and estimation, custom allocators, and index threadpool size control (#305, #306, #352, #354, #358, #363, #364, #370, #380)
 
-* Clang-21 compiler support with gcc-11 headers for shared library builds
+* Experimental fine-grain concurrent Vamana index, `svs::index::vamana::concurrent::MutableVamanaIndex`, supporting lock-free search concurrent with `add_points`, `delete_entries`, and `consolidate` (#369)
 
-* Amazonlinux2 support for shared/static library dropped
+* LVQ and LeanVec dataset support for the fine-grain concurrent Vamana index
 
-* Consolidated to publish only minimal necessary shared library (14 -> 6) and runtime bindings (2 -> 1) release artifacts
+* `replace_external_id` added to `MutableVamanaIndex` to rename a vector's external ID without re-inserting its data (#383)
 
-* Adaptive batch size heuristic for filtered search in Vamana, with new `filter_stop` early-exit parameter (#309)
+* `get_memory_usage()` added to VamanaIndex to report allocated bytes (#345)
 
-* IVF runtime support for `IDFilter`-based filtered search and dynamic `set_intra_query_threads` adjustment after `train()`/`load()` (#322)
+* Optional `blocksize_elements` added to `BlockingParameters` (#344)
 
-* IVF inter-query thread pool initialization now validates and clamps thread count against the number of centroids (#329)
+* `element_size()` added to LVQ and LeanVec datasets
 
-* In-memory (mmap) stream index mapping with new `map_to_file` and `map_to_memory` APIs on `FlatIndex` and `VamanaIndex` (#310)
+* `CompressedDataset` instantiations added to the shared library
 
-* Add get_distance and reconstruct_at to VamanaIndex API (#315)
+* Reduced peak memory of LeanVec index builds by sizing transform batches automatically
+
+* Fixed LeanVec `is_pca` to derive centering from matrix equality rather than matrix presence
+
+* Fixed GCC-12.x prefetch-loop collapse in `greedy_search` neighbor prefetch (#361)
+
+* `Blocked` class refactored to meet allocator requirements (#351)
+
+* `search_buffer_visited_set_` is now `true` by default for increased performance (#369)

bindings/c/CMakeLists.txt

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 cmake_minimum_required(VERSION 3.21)
-project(svs_c_api VERSION 0.4.0 LANGUAGES CXX C)
+project(svs_c_api VERSION 0.5.0 LANGUAGES CXX C)
 set(TARGET_NAME svs_c_api)
 
 include(GNUInstallDirs)

bindings/cpp/CMakeLists.txt

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 cmake_minimum_required(VERSION 3.21)
-project(svs_runtime VERSION 0.4.0 LANGUAGES CXX)
+project(svs_runtime VERSION 0.5.0 LANGUAGES CXX)
 set(TARGET_NAME svs_runtime)
 
 # IVF requires MKL, so it's optional

bindings/cpp/conda-recipe/meta.yaml

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 
-{% set version = '0.4.0' %}
+{% set version = '0.5.0' %}
 {% set buildnumber = 0 %}
 {% set variant_suffix = environ.get('SUFFIX', '').replace('-', '_') %}
 

bindings/cpp/include/svs/runtime/version.h

@@ -41,7 +41,7 @@
 
 #ifndef SVS_RUNTIME_VERSION_MINOR
 /// Minor version number - incremented for backward-compatible feature additions
-#define SVS_RUNTIME_VERSION_MINOR 4
+#define SVS_RUNTIME_VERSION_MINOR 5
 #endif
 
 #ifndef SVS_RUNTIME_VERSION_PATCH
@@ -51,7 +51,7 @@
 
 #ifndef SVS_RUNTIME_VERSION_STRING
 /// Complete version string
-#define SVS_RUNTIME_VERSION_STRING "0.4.0"
+#define SVS_RUNTIME_VERSION_STRING "0.5.0"
 #endif
 
 #ifndef SVS_RUNTIME_API_VERSION

bindings/python/setup.py

@@ -40,7 +40,7 @@
 
 setup(
     name="scalable-vs",
-    version="0.4.0",
+    version="0.5.0",
     description="Scalable Vector Search (SVS) is a performance library for vector similarity search.",
     long_description=long_description,
     long_description_content_type="text/markdown",

bindings/python/tests/test_common.py

@@ -42,7 +42,7 @@ def tearDown(self):
     #####
 
     def test_version(self):
-        self.assertEqual(svs.library_version(), "v0.4.0")
+        self.assertEqual(svs.library_version(), "v0.5.0")
 
     def test_conversion(self):
         # signed

tests/svs/lib/version.cpp

@@ -46,6 +46,6 @@ CATCH_TEST_CASE("Version Numbers", "[lib][versions]") {
 
 // Keep in-sync with CMakeLists.txt
 CATCH_TEST_CASE("Global Version", "[lib][versions]") {
-    static_assert(svs::lib::svs_version == svs::lib::Version(0, 4, 0), "Version mismatch!");
-    CATCH_REQUIRE(svs::lib::svs_version == svs::lib::Version(0, 4, 0));
+    static_assert(svs::lib::svs_version == svs::lib::Version(0, 5, 0), "Version mismatch!");
+    CATCH_REQUIRE(svs::lib::svs_version == svs::lib::Version(0, 5, 0));
 }