Adrian Wälchli

In the Google Chrome console I see the following error: `Uncaught SyntaxError: Unexpected token 'if'` The error appears near ` if(g>z){break;}}}}` in the minified js file. However, I can't find...

Fix device placement when `.cuda()` called without specifying index

3

## What does this PR do? The PR #13704 resulted in an unintended change. The second `.cuda()` call below should leave the model on the device set by `torch.cuda.set_device` if...

See title ## Before submitting - [x] Was this discussed/approved via a GitHub issue? (not for typos and docs) - [x] Did you read the [contributor guideline](https://github.com/PyTorchLightning/pytorch-lightning/blob/master/.github/CONTRIBUTING.md), **Pull Request** section?...

Replace unwrapping logic in strategies

2

## What does this PR do? Fixes #13502 (main RFC) Fixes #8593 Fixes #13500 The PR looks big, but most complexity turns out to be in the deprecation. The rest...

Convert subprocess test to standalone test

1

## What does this PR do? Needed for #13967 In #13967 there are two tests that [fail](https://dev.azure.com/Lightning-AI/lightning/_build/results?buildId=86941&view=logs&j=3f274fac-2e11-54ca-487e-194c91f3ae9f&t=8da0b6c4-35de-5c68-3c64-0e4ce0eb560c&l=93): - test_multi_gpu_model_ddp_fit_only - test_multi_gpu_model_ddp_fit_test These tests are unrelated to deepspeed, yet they fail...

Patch Release v1.7.2 ``` gh pr list -s merged -S 'merged:2022-08-09T20:30:00.000Z..2022-08-16T22:00:00.000Z' --json author,milestone --jq '.[] | select((.milestone.title == "pl:1.7.x") or (.milestone.title == null)) | ["@" + .author.login] | join("")' --limit...

Update tests that mock cuda availability and extend it for mps

2

## Proposed refactor We have tests that fail locally on MPS supported devices because their mocking of CUDA does not extend to MPS. After #13947 landed, the availability check actually...

RFC: Remove `num_nodes` Trainer argument and infer world size from cluster environment directly

4

## 🚀 Feature Remove the redundant `num_nodes` Trainer argument. Knowing the number of nodes is not required, and the world size is provided by the cluster environment anyway. ### Motivation...

RFC: Make subprocess DDP the default when selecting multiple devices

3

## 🚀 Feature RFC: Switch to strategy="ddp" when selecting devices>1. I'm putting this RFC out so it can be discussed. There is no urgency to this and we don't know...

## What does this PR do? Recently DeepSpeed resolved https://github.com/microsoft/DeepSpeed/issues/2139 which will make their latest release again compatible with LightningLite. Hence we are removing the previously added user errors. This...