Skip batch norm when chunk size is 1 in GhostBatchNorm

[arnavgarg1]

Causes the following error otherwise:

ValueError: Expected more than 1 value per channel when training, got input size torch.Size([1, 6])

[tgaddair]

Nice! Only other thing I might consider is logging a warning for each of the two cases where we skip the norm:

• batch_size <= virtual_batch_size
• batch_size % virtual_batch_size == 1

[arnavgarg1]

Nice! Only other thing I might consider is logging a warning for each of the two cases where we skip the norm:

• batch_size <= virtual_batch_size
• batch_size % virtual_batch_size == 1

Added! Let me know if this makes sense

[github-actions]

Unit Test Results

         6 files  ±  0           6 suites  ±0   5h 51m 3s ⏱️ - 9m 46s
  3 968 tests +12    3 931 ✔️ +12    37 💤 ±0  0 ❌ ±0 
11 901 runs  +77  11 790 ✔️ +76  111 💤 +1  0 ❌ ±0 

Results for commit 9dc3b8a. ± Comparison against base commit c638c23.