A distributed approximate nearest neighborhood search (ANN) library which provides a high quality vector index build, search and distributed online serving toolkits for large scale vector search scenario.
Restructure TPT construction when using PQ or OPQ quantization to fix issue where graph accuracy was low. Now achieves CPU level of accuracy when sufficient TPTrees are used.
Outstanding issues:
Need more testing of OPQ support for GPU buildHead and buildSSD
Performance of index build with PQ/OPQ is quite slow due to distance table lookups, more work needed to optimize
Batching of reconstructed vectors needed during TPT construction to enable PQ/OPQ build of larger datasets.
Restructure TPT construction when using PQ or OPQ quantization to fix issue where graph accuracy was low. Now achieves CPU level of accuracy when sufficient TPTrees are used.
Outstanding issues: