TQDM Progress Bar shows negative time and speed after resuming from checkpoint

[ductai199x]

🐛 Bug

TQDM Progress Bar shows negative time and speed after resuming from checkpoint

Screencast.from.07-01-2022.10.12.29.PM.webm

To Reproduce

Train any model for sometime, then resume. The progress bar will register negative numbers

Expected behavior

Resuming should NOT cause the progress bar to display negative ETA/speed

Environment

• CUDA:
  - GPU:
  - NVIDIA GeForce RTX 3090
  - available: True
  - version: 11.6
• Packages:
  - numpy: 1.23.0
  - pyTorch_debug: False
  - pyTorch_version: 1.12.0+cu116
  - pytorch-lightning: 1.6.4
  - tqdm: 4.64.0
• System:
  - OS: Linux
  - architecture:
  - 64bit
  - ELF
  - processor: x86_64
  - python: 3.9.13
  - version: fixed clip grad warning #42-Ubuntu SMP Thu Jun 9 23:42:32 UTC 2022

[awaelchli]

This is a duplicate of #13124. I'm closing this to keep the discussion in #13124. Feel free to subscribe to the issue to get notified about the progress.

I had already looked into it but couldn't find the reason why the bar shows the incorrect estimates yet.