Tamas Bela Feher

```bash python -m raft-ann-bench.generate_dataset --rows 1000000 --cols 128 --dtype float32 dataset/base.fbin # After the dataset is generated, you can create query and ground truth files python -m raft-ann-bench.generate_groundtruth dataset/base.fbin --output=dataset...

Note this will only compile once #2155 is merged.

Thanks Artem for proposing this solution. On one hand, it is nice to have a secondary workspace allocator to handle large allocations. I need to still think about this. An...

Thanks Artem for the update! It is a nice idea to have an extra memory resource that we can use for potentially host mem backed large temporary allocations. This can...

Tagging @abc99lr who plans to work on this, and @achirkin for visibility.

This issue shall be implemented as a follow up to #2169, because that PR changes a few details in the IVF-Flat fine search.

Hi @vilmara, thanks for reporting the error. Normally TF-TRT should exclude string nodes from the conversion, and convert the rest of the model. The error message suggests that there is...

The k-means clustering implementation calls [predict](https://github.com/rapidsai/raft/blob/branch-23.12/cpp/include/raft/cluster/detail/kmeans_balanced.cuh#L672) multiple times during its iteration. The [FusedL2NN](https://github.com/rapidsai/raft/blob/637019e676831c3d38fde4c6e3dfbcae066e1c23/cpp/include/raft/cluster/detail/kmeans_balanced.cuh#L117) operation is used in predict, which under the hood would call [CUTLASS](https://github.com/rapidsai/raft/blob/branch-23.12/cpp/include/raft/distance/detail/fused_l2_nn.cuh#L347) to enable 3xTF32 computations. This...

>If we find that 1xTF32 path is enough in terms of accuracy and is better performing, we will keep 1xTF32 as a default path for FP32 ANN instead of current...

Thanks for the detailed report. It is a valid use case to have multiple models with multiple pre-built engines. We seem to have a problem with the way the engines...