How to use faiss.index_cpu_to_all_gpus() for faiss.IndexBinaryFlat?

[anoubhav]

I want to use multiple GPUs while using the binary flat index.
When I run faiss.index_cpu_to_all_gpus(faiss.IndexBinaryFlat(d)), I get the following error:

TypeError: Wrong number or type of arguments for overloaded function 'index_cpu_to_gpu_multiple'.
  Possible C/C++ prototypes are:
    faiss::gpu::index_cpu_to_gpu_multiple(std::vector< faiss::gpu::GpuResourcesProvider * > &,std::vector< int > &,faiss::Index const *,faiss::gpu::GpuMultipleClonerOptions const *)
    faiss::gpu::index_cpu_to_gpu_multiple(std::vector< faiss::gpu::GpuResourcesProvider * > &,std::vector< int > &,faiss::Index const *)

I don't get this error when using L2 flat index. faiss.index_cpu_to_all_gpus(faiss.IndexFlatL2(d))

[mdouze]

faiss.index_cpu_to_all_gpus is only for float indexes.
For binary indexes (whether CPU or GPU), the construction should be done manually.
See here for an example on how to do that:

faiss/tests/test_index_binary.py

Line 341 in 7492d23

class TestReplicasAndShards(unittest.TestCase):

[anoubhav]

Just to clarify: "construction should be done manually" means that I cannot train a binary index with a GPU compatible binary clustering index?

https://gist.github.com/mdouze/dd11f1ebd2f1c2f3bcd74beee303e513
In the above link, you shared how to use a GPU-compatible faiss.IndexFlatL2 to build a binary IVF index.

Is it possible to use GPU-optimized faiss.IndexBinaryFlat to build a binary IVF index?
Thanks,

[anoubhav]

Is it possible to perform a search using faiss.IndexBinaryFlat on multiple GPUs?

Using faiss.GpuIndexBinaryFlat I am able to do the search on a single GPU. But not on more than 1 GPUs due to the above error when trying the binary flat index in faiss.index_cpu_to_all_gpus