DeMoriarty

Hi, Cupy is a core part of the library for JIT-compiling cuda kernels.

Hi, I could not reproduce your results exactly, even with 10000 parallel kmeans, I'm getting: ``` Counter({50: 9995, 49: 5}) ```

Can you confirm you are using the exact same code you provided in your first comment? if that's the case I'm not sure what could cause this to happen. I'm...

Hi, this is currently not possible with TorchPQ.

嗨，我想这可能是你要找的 [https://github.com/DeMoriarty/TorchPQ/blob/main/torchpq/codec/PQCodec.py](https://github.com/DeMoriarty/TorchPQ/blob/main/torchpq/codec/PQCodec.py) ```python from torchpq.codec import PQCodec codec = PQCodec(d_vector, n_subvectors, n_clusters) d_vector = 128 # vector dimension n_vectors = 10000 # number of vectors n_subvectors = 8 # number...

please make sure you have nvcc installed on your system, run: ```bash which nvcc ``` on linuxs, or: ```bash where nvcc ``` on windows. it should return the path of...

if `/home/oushou/cuda-11.0/` is where cuda is installed, then you should use the nvcc in `/home/oushou/cuda-11.0/bin/`. also make sure the installed version of cupy is `cuda-cupy110`

the kernel in master branch is an older version. I have fixed the bug in the main branch, can you try again?

the performance of master branch isn't good, because its an older version of the bmm kernel, which is not as optmized as the kernel in the main branch. I have...

Yes, as I explained in [this blog post ](https://demoriarty.github.io/BMM-1/), this kernel is optimized for Turing series GPUs (such as Tesla T4, RTX 2080, Titan RTX...). For better performance on Ampere...