Customer Facing Features

Training:

• multi-GPU training
  • extend current train.py with multi-GPU capability
    • add profile information (nccl)

Loaders (100% code re-use):

• segy loaders/converters (in case customers ask about them)
  • merge contrib from CSE into staging
    • segy to numpy format converter utility (already have from CSE)
    • segy notebooks
    • generic converter for Dutch F3 (tensor format) dataset (cannot convert some datasets to segy) [stretch goal]
    • with unit tests
• specific converter for Penobscot dataset
  - bring back Penobscot: convert to numpy, import through Dutch F3 interface

Enhancements:

• speed up build process
  - overhaul tests