Description
TorchServe has a varying structure for different accelerator types through recently added #3371.
Although Rebellions provides a guide on how to utilize TorchServe with the RBLN(Rebellions) NPUs
through its official document page(https://docs.rbln.ai/software/model_serving/torchserve/torchserve.html), the current implementation of TorchServe does not recognize the RBLN NPU as a valid accelerator vendor. As a result, even when gpu_id
is set in configuration using the RBLN NPU
, the specified RBLN NPUs cannot be properly utilized.
We would like to propose adding RBLN NPU as a recognized accelerator vendor in TorchServe, along with an official user guide. This addition will enable seamless integration and usage of TorchServe in environments equipped with RBLN NPUs.