Introspection | USearch for Python

index.count(42)  # entries sharing this key
42 in index  # the same question, as a membership test
len(index)  # stored entries
index.describe()  # a DenseIndexDescription: .metadata, .size, .capacity, .memory_usage, .cpu_class, .gpu_class, .shards_count
index.health()  # a DenseReachability: .audited, .reachable, .sweeps
index.cpu_class  # the CPU class dispatch picked on this machine
usearch.capabilities_compiled()  # the kernel families this build carries
usearch.capabilities_available()  # the ones this machine can run

capacity, dimensions, connectivity and memory_usage are properties read through describe as well. health audits how much of the graph a search reaches, and the entries it cannot reach wait for a refine to re-link them. Every docstring names the C symbol it calls, or says the method is Python-only.

API

Types

CPUClassThe CPU families dispatch picks from; a placement may pin one.

__init__.py:237 ↗

class CPUClass(IntEnum)

Values

`SERIAL`	`1`	The portable floor: no SIMD assumptions.
`INTEL_HASWELL`	`2`	Haswell-class x86: AVX2 and FMA, the 256-bit floor.
`INTEL_SIERRA`	`3`	Sierra Forest-class x86, the E-core Xeons: AVX-VNNI without AVX-512.
`INTEL_SKYLAKE`	`4`	Skylake-class x86: the AVX-512 foundation tier.
`INTEL_ICELAKE`	`5`	Icelake-class x86: AVX-512 VNNI and VPOPCNT.
`AMD_GENOA`	`6`	Genoa-class x86, Zen 4 and newer: native bf16 dot products over AVX-512.
`INTEL_SAPPHIRE`	`7`	Sapphire Rapids-class x86: the Genoa bundle plus AVX-512 FP16 and AMX tiles.
`INTEL_GRANITE`	`8`	Granite Rapids-class x86: the Sapphire Rapids bundle plus AMX-FP16 tiles.
`INTEL_DIAMOND`	`9`	Diamond Rapids-class x86: the Granite Rapids bundle plus AVX10.2 FP8 conversions.
`AWS_G2`	`10`	Graviton2-class Arm, Neoverse N1: the bare NEON floor.
`APPLE_M2`	`11`	Armv8.6 NEON with dot products and bf16 and f16 fused multiplies, as in Apple M2.
`APPLE_M4`	`12`	Apple M4-class Arm: the Armv8.6 NEON bundle plus SME outer-product tiles.
`AWS_G5`	`13`	Graviton5-class Arm, Neoverse V3 and Armv9.2: monitored waits.
`NVIDIA_VERA`	`14`	Vera-class Arm: NVIDIA's Armv9.2-generation cores with FP8 dot products.
`ALIBABA_C930`	`15`	Alibaba C930-class RISC-V, RVA23: vectors with Zvbb, monitored waits.
`WASM32_V128`	`16`	WebAssembly every engine loads: 32-bit, single-threaded, simd128.
`WASM64_EXTENDED`	`17`	WebAssembly with relaxed SIMD, shared memory and 64-bit addressing.
`MONOVERSIONED`	`18`	The one class of a monoversioned build, carrying every capability its floor implies.

GPUClassThe GPU families dispatch picks from; a placement may pin one.

__init__.py:277 ↗

class GPUClass(IntEnum)

Values

`NVIDIA_AMPERE`	`1`	Ampere-class NVIDIA and every NVIDIA GPU without a class of its own: A100, A10, RTX 30.
`NVIDIA_ADA`	`2`	Ada-class NVIDIA: L4, L40S and RTX 40, with fp8 conversions.
`NVIDIA_HOPPER`	`3`	Hopper-class NVIDIA: H100 and H200, with DPX instructions.
`NVIDIA_BLACKWELL`	`4`	Blackwell-class NVIDIA: B200, B300 and GB300, with TMA gather4 fetches.
`NVIDIA_BLACKWELLRTX`	`5`	Consumer Blackwell NVIDIA: RTX 50, RTX PRO 6000 and DGX Spark.
`NVIDIA_RUBIN`	`6`	Rubin-class NVIDIA, the Vera-Rubin platform's device half.
`AMD_CDNA3`	`7`	CDNA3 AMD: MI300X and MI325X, 64-wide waves, FNUZ fp8.
`AMD_CDNA4`	`8`	CDNA4 AMD: MI350X and MI355X, OCP fp8.
`AMD_CDNA5`	`9`	CDNA5 AMD: MI400, 32-wide waves.
`APPLE_METAL7`	`10`	Apple7-family GPUs and every Apple GPU without a class of its own: M1 and M2.
`APPLE_METAL9`	`11`	Apple9-family GPUs: M3 and M4.
`APPLE_METAL10`	`12`	Apple10-family GPUs: M5, with Neural Accelerator matrix units.

NVMeClassThe I/O engines NVMe executors submit through; an executor may pin one.

__init__.py:305 ↗

class NVMeClass(IntEnum)

Values

`POSIX`	`1`	Blocking reads on the calling thread.
`LINUX_AIO`	`2`	Linux AIO with direct I/O.
`LINUX_URING`	`3`	io_uring reads on files and block devices.
`LINUX_URING_PASSTHROUGH`	`4`	io_uring NVMe passthrough on namespace character devices, such as "/dev/ng0n1".
`LINUX_SPDK`	`5`	The SPDK user-space driver, over namespaces named by transport identifiers.
`WINDOWS_IORING`	`6`	I/O Rings, on Windows 11 22H2 and newer.

ClassDenseIndexDescriptionA live index's frozen metadata and current counters.7

lib.pyi:1692 ↗

class DenseIndexDescription(structseq[Any], tuple[DenseMetadata, int, int, int, CPUClass | None, GPUClass | None, int])

metadataThe frozen DenseMetadata, as a checkpoint reports it.

lib.pyi:1707 ↗

metadata: DenseMetadata

sizeStored entries.

lib.pyi:1712 ↗

size: int

capacityEntries hosted without growing again.

lib.pyi:1717 ↗

capacity: int

memory_usageBytes the handle holds; traversal scratch is the executor's.

lib.pyi:1722 ↗

memory_usage: int

cpu_classThe CPUClass dispatch picked on this machine, or None.

lib.pyi:1727 ↗

cpu_class: CPUClass | None

gpu_classThe GPUClass dispatch picked on the index's devices, or None for a host index.

lib.pyi:1732 ↗

gpu_class: GPUClass | None

shards_countLive shards.

lib.pyi:1737 ↗

shards_count: int

ClassDenseReachabilityOne reachability audit.4

lib.pyi:1741 ↗

class DenseReachability(structseq[Any], tuple[int, int, int, int])

auditedEntries published when the audit started, the population it judged.

lib.pyi:1748 ↗

audited: int

reachableOf those, the entries a search reaches; the rest wait for a refine to re-link them.

lib.pyi:1753 ↗

reachable: int

sweepsLevels the sweep took to converge.

lib.pyi:1758 ↗

sweeps: int

back_sweepsLevels the sweep back toward the entry point took, where an audit sweeps both ways.

lib.pyi:1763 ↗

back_sweeps: int

Functions

versionThe library's semantic version.

__init__.py:79 ↗

def version() -> str

capabilities_compiledThe kernel families this build carries.

__init__.py:83 ↗

def capabilities_compiled() -> str

capabilities_availableThe kernel families this machine can run.

__init__.py:87 ↗

def capabilities_available() -> str

layoutsThe size and field offsets of every argument struct the other bindings lay out by hand.

__init__.py:91 ↗

def layouts() -> dict[str, tuple[int, dict[str, int]]]