This page describes the index build parameters and index search parameters for the CAGRA algorithm.
Index build parameters
| Parameter | Type | Default value | Description |
|---|---|---|---|
proxima.cagra.index.graph_degree | int | 64 | Number of edges per node in the nearest neighbor graph during queries. Higher values improve recall but reduce throughput. |
proxima.cagra.index.intermediate_graph_degree | int | 128 | Number of edges per node in the nearest neighbor graph during the graph creation phase. The graph is pruned down to graph_degree at query time. Must be greater than or equal to graph_degree and no more than 1.5x graph_degree. |
proxima.cagra.index.mode.fp16 | int | 0 | Specifies whether to enable FP16 quantization. Set to 1 to enable. Enabling FP16 mode is recommended to avoid degrading recall. |
Index search parameters
| Parameter | Type | Default value | Description |
|---|---|---|---|
proxima.cagra.search.itopk_size | int | 64 | Size of the internal candidate set during each iteration. This is the primary parameter for tuning the recall-throughput trade-off. Must be a multiple of 32 and greater than or equal to your target top-k value. For example, set to at least 100 when retrieving top-100 results. Higher values improve recall but reduce throughput. |
proxima.cagra.search.use_batchable_searcher | int | 0 | Specifies whether to enable batch processing. Set to 1 to enable batch processing. This way, the system can process query requests that arrive within a short time window in a batch, which significantly improves throughput. |
proxima.cagra.search.use_gpu_built_index | int | 0 | Specifies the index source. Only 0 is supported, which uses the graph built by the CPU. |
该文章对您有帮助吗？