Jinjing Zhou

Made an issue at https://github.com/tensorchord/pgvecto.rs/issues/503

This is the expected behavior, that all the int32 tensors are on cpu by default, no matter which context device is specified.

You can force it on gpu by using `tf.identity` operator

@miguelusque I'm not tf developer, but met the same issue before. Generally speaking this is a legacy issue from tensorflow 1. int32 tensors were usually used to represent the shape...

Some issues related to this https://github.com/tensorflow/tensorflow/issues/34071, https://github.com/tensorflow/tensorflow/issues/41307

- Identify the precomputation part - Implement the distance for L2 square - Implement the distance for dot product and cosine distance - Implement fast scan distance computation

This usually has nothing to do with pgvecto.rs. It's related to the original folder permission and the user id used in the docker image. We need more information about the...

@cjnolet Thanks, we're developing a vector database with a kind of quantization with IVF index. We need the KMeans results first and then do the quantization part after that. And...

@cjnolet We only need the centroids from the GPU KMeans part. The assignment and quantization part will be done at the database side with CPU due to data consistency issue...

@cjnolet Thanks. In terms of my question, I think the better way to solve it would be kind of mini batch KMeans algorithm, or leverage the CPU memory by cuda...