README.md multi-GPU training instructions add bold emphasis to how numGPUs influences batch size

[mkrupczak3]

I found this description to be incredibly subtle, I ran a lot of training runs with the wrong batch size before I saw it

Description

Simple change to README.md to add bold emphasis under Multi-GPU training instructions:

Note that your effective batch size is multiplied by the number of GPUs**, so you may need to adjust your batch_size and grad_accum_steps to maintain the same overall batch size!

No dependencies as it is just a documentation update

Type of change

Please delete options that are not relevant.

• This change requires a documentation update

How has this change been tested, please provide a testcase or example of how you tested the change?

With my eyes reading the README.md

Any specific deployment considerations

May be worth emphasizing in documentation elsewhere to make sure others don't make the same mistake that I did

Docs

• Docs updated? What were the changes:
  Just a simple change to README.md

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}