Problem: Pipeline Parallelism with distributed-rpc on Jetson Nano devices

distributed distributed-rpc
1

Hi, i’m a student and currently working on pipelining models on edge devices.

My current setup is: 1 Master linux machine, 2 Jetson Nano as workers.
I used the example on:
https://pytorch.org/tutorials/intermediate/dist_pipeline_parallel_tutorial.html

I also trying to investigate the impact of split size (for parallel execution) on the performance.

I have couple of queries as I have problem connecting to the edge devices using their IP addresses. And later I also want to do the performance analysis for each split using Pytorch.

2

Hi, did you have solution. Now i have one project like yours, and unable to solve it.

  1. how to split the model, such as Transformer
  2. how to communicate between the master machine and worker

Please give some advices. Thanks.