diff --git a/site/en/reference/index.md b/site/en/reference/index.md index c8c85950d..2c8606a44 100644 --- a/site/en/reference/index.md +++ b/site/en/reference/index.md @@ -409,15 +409,16 @@ Combined with SQ, HNSW_SQ offers a controllable trade-off between index size and | `M` | M defines tha maximum number of outgoing connections in the graph. Higher M leads to higher accuracy/run_time at fixed ef/efConstruction. | [2, 2048] | None | | `efConstruction` | ef_construction controls index search speed/build speed tradeoff. Increasing the efConstruction parameter may enhance index quality, but it also tends to lengthen the indexing time. | [1, int_max] | None | | `sq_type` | Scalar quantizer type. | `SQ6`,`SQ8`, `BF16`, `FP16` | `SQ8` | + | `refine` | Whether refined data is reserved during index building. | `true`, `false` | `false` | + | `refine_type` | The data type of the refine index. | `SQ6`, `SQ8`, `BF16`, `FP16`, `FP32` | None | - Search parameters | Parameter | Description | Range | Default Value | |---------------|------------------------------------------------------------------------------------------------------------|--------------------------------------|---------------| | `ef` | Parameter controlling query time/accuracy trade-off. Higher `ef` leads to more accurate but slower search. | [`top_k`, int_max] | None | - | `refine` | Whether the refine is used during the train. | `true`, `false` | `false` | | `refine_k` | The magnification factor of refine compared to *k*. | [1, *float_max*) | `1` | - | `refine_type` | The data type of the refine index. | `SQ6`, `SQ8`, `BF16`, `FP16`, `FP32` | None | + ### HNSW_PQ @@ -433,15 +434,15 @@ Combined with PQ, HNSW_PQ offers a controllable tradeoff between index size and | `efConstruction` | ef_construction controls index search speed/build speed tradeoff. Increasing the efConstruction parameter may enhance index quality, but it also tends to lengthen the indexing time. | [1, int_max] | None | | `m` | The number of sub-vector groups to split the vector into. | [1, 65536] | 32 | | `nbits` | The number of bits into which each group of sub-vectors is quantized. | [1, 24] | 8 | + | `refine` | Whether refined data is reserved during index building. | `true`, `false` | `false` | + | `refine_type` | The data type of the refine index. | `SQ6`, `SQ8`, `BF16`, `FP16`, `FP32` | None | - Search parameters | Parameter | Description | Range | Default Value | |---------------|------------------------------------------------------------------------------------------------------------|--------------------------------------|---------------| | `ef` | Parameter controlling query time/accuracy trade-off. Higher `ef` leads to more accurate but slower search. | [`top_k`, int_max] | None | - | `refine` | Whether the refine is used during the train. | `true`, `false` | `false` | | `refine_k` | The magnification factor of refine compared to *k*. | [1, *float_max*) | `1` | - | `refine_type` | The data type of the refine index. | `SQ6`, `SQ8`, `BF16`, `FP16`, `FP32` | None | ### HNSW_PRQ @@ -458,15 +459,15 @@ Combined with a Product Residual Quantizer (PRQ), HNSW_PRQ offers an even higher | `m` | The number of sub-vector groups to split the vector into. | [1, 65536] | 32 | | `nbits` | The number of bits into which each group of sub-vectors is quantized. | [1, 24] | 8 | | `nrq` | The number of residual subquantizers. | [1, 16] | 2 | + | `refine` | Whether refined data is reserved during index building. | `true`, `false` | `false` | + | `refine_type` | The data type of the refine index. | `SQ6`, `SQ8`, `BF16`, `FP16`, `FP32` | None | - Search parameters | Parameter | Description | Range | Default Value | |---------------|------------------------------------------------------------------------------------------------------------|--------------------------------------|---------------| | `ef` | Parameter controlling query time/accuracy trade-off. Higher `ef` leads to more accurate but slower search. | [`top_k`, int_max] | None | - | `refine` | Whether the refine is used during the train. | `true`, `false` | `false` | | `refine_k` | The magnification factor of refine compared to *k*. | [1, *float_max*) | `1` | - | `refine_type` | The data type of the refine index. | `SQ6`, `SQ8`, `BF16`, `FP16`, `FP32` | None |