benchmark icon indicating copy to clipboard operation
benchmark copied to clipboard
verified publisher icon pytorch

Enhance TorchBench coverage for large distributed workloads with CI support for IBM Cloud

Open spzala opened this issue 2 years ago • 1 comments

The proposed work tasks are as below:

  • [ ] Enable CI support for IBM Cloud to enhance the testing infrastructure for FSDP
  • [ ] Benchmark new model(s) for FSDP training - e.g. add new hf_T5 with 3B parameters, etc.

spzala avatar Apr 10 '23 14:04 spzala

Looping @mrshenli into the discussion, since the PyTorch Distributed team is also interested in building benchmarks in distributed setup.

xuzhao9 avatar Apr 11 '23 17:04 xuzhao9

We tested the workflow. Closing it for now and will reopen when we decide to move further.

spzala avatar Jul 07 '25 15:07 spzala