Rohit Gupta

Fix val dataloader reload conditions when `check_val_every_n_epochs` is set

1

## What does this PR do? Fixes #12145 ### Does your PR introduce any breaking changes? If yes, please list them. ## Before submitting - [x] Was this **discussed/approved** via...

Avoid raising the sampler warning if num_replicas=1

2

## What does this PR do? Dont' raise a warning if devices=1 already and ddp is still used. ### Does your PR introduce any breaking changes? If yes, please list...

Use fsdp module to initialize precision scalar for fsdp native

1

## What does this PR do? Fairscale shouldn't be a requirement when using fsdp_native, so we should rely on fsdp package from torch itself. ### Does your PR introduce any...

## What does this PR do? Added docs for `fsdp_native`. ### Does your PR introduce any breaking changes? If yes, please list them. ## Before submitting - [ ] Was...

Profile batch transfer and gradient clipping hooks

2

## What does this PR do? Fixes #14028 ### Does your PR introduce any breaking changes? If yes, please list them. ## Before submitting - [x] Was this **discussed/approved** via...

## Proposed refactor ### Motivation Currently, batch transfer hooks are disabled for IPUs and DP strategy for very valid reasons. But call to `on_before_batch_transfer` shouldn't be limited to whether any...

## What does this PR do? Call to `on_before_batch_transfer` shouldn't be limited to whether any strategy/accelerator supports explicit transfer to batches to the device. Users should still be able to...

Add BatchSizeFinder callback

5

## What does this PR do? Part of #11012 TODO: - Add docs in followups - Address added TODOs here in the followup ### Does your PR introduce any breaking...

source links are broken in docs

1

## 📚 Documentation There's something wrong with metrics source redirect: 1. On `stable`, If you click on source, it doesn't show up anything on master which is fine I guess...