Full Retrieval

Full Retrieval Algorithms

Algorithm (API Reference)	Application Scenario	Operators to Be Generated	Example
AscendIndexInt8Flat	Feature type: int8 Feature dimension: 64, 128, 256, 384, 512, 768, or 1024 Distance type: L2 or IP Calculation precision: high Device memory usage: low Application scenario: brute-force search with high precision requirements	INT8Flat AICPU	Link
AscendIndexFlat	Feature type: FP32 or FP16 Feature dimension: 32, 64, 128, 256, 384, 512, 768, 1024, 1408, 1536, 2048, 3072, 3584, or 4096 Distance type: L2 or IP Calculation precision: high Device memory usage: high Application scenario: brute-force search with high precision requirements. It is recommended that the IP distance be used in scenarios where dim is greater than 128.	Flat AICPU	Link
AscendIndexSQ	Feature type: FP32 Feature dimension: 64, 128, 256, 384, 512, or 768 Distance type: L2 or IP Calculation precision: high Device memory usage: low (quantized to int8) Application scenario: brute-force search with high precision requirements	SQ8 AICPU	Link
AscendIndexCluster	Feature type: FP32 Feature dimension: 32, 64, 128, 256, 384, or 512 Distance type: IP Calculation precision: high Device memory usage: high Application scenario: clustering scenario where only the distance is calculated It is supported only by the Atlas inference product.	Flat AICPU	Link
IndexIL	(Not recommended) It needs to run on the device, and the installation and deployment are complex.	Flat	IndexILFlat
AscendIndexILFlat	Feature type: FP16 or FP32 Feature dimension: 32, 64, 128, 256, 384, or 512 Distance type: IP Calculation precision: high Device memory usage: high Application scenario: clustering scenario where only the distance is calculated It is supported only by the Atlas inference product.	Flat AICPU	Link

Parent topic: Algorithm Introduction