We are researchers at a Public University in the USA. We have developed a new variant of the DLPack protocol to make it easy for consumers to zero-copy share the deep learning framework's tensor (e.g., PyTorch, TensorFlow, etc.) to any consumer system. It removes the need to have consumer APIs. More details are available in our research paper (https://arxiv.org/pdf/2404.04118, section IV-A).
There are two things we are expecting from this GitHub issue:
We would be happy to incorporate any feedback that you may provide about this new version of the protocol. We call it producer-only DLPack protocol. For reference, the current protocol is labeled as the producer-consumer DLPack protocol.
Hi,
We are researchers at a Public University in the USA. We have developed a new variant of the DLPack protocol to make it easy for consumers to zero-copy share the deep learning framework's tensor (e.g., PyTorch, TensorFlow, etc.) to any consumer system. It removes the need to have consumer APIs. More details are available in our research paper (https://arxiv.org/pdf/2404.04118, section IV-A).
There are two things we are expecting from this GitHub issue:
We would be happy to incorporate any feedback that you may provide about this new version of the protocol. We call it producer-only DLPack protocol. For reference, the current protocol is labeled as the producer-consumer DLPack protocol.