Yifu Wang

make allocators and sanitizers work for processes created with multiprocessing's spawn method in dev mode

6

Summary: **The first attempt (D30802446) overlooked that fact that the interpreter wrapper is not an executable (for `execv`) on Mac, and introduced some bugs due to the refactoring. The attempt...

make allocators and sanitizers work for processes created with multiprocessing's spawn method in dev mode

3

Summary: #### Problem Currently, the entrypoint for in-place Python binaries (i.e. built with dev mode) executes the following steps to load system native dependencies (e.g. sanitizers and allocators): - Backup...

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #67 * #66 When the installed PyTorch version is not nightly: - Issue an warning if `torch.compile` is requested. - Raise an...

[dynamo] support group=None when rewriting collectives

5

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #120560 * #120181 * __->__ #120118 * #120370 Resolves case 2 in #120082. cc @mrshenli @pritamdamania87 @zhaojuanmao @satgera @rohan-varma @gqchen @aazzolini @osalpekar @jiayisuse...

Prepare test_inductor_collectives.py for native funcol migration

4

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #119951 * #120118 * #120070 * #120043 * #120042 * __->__ #120025 * #119982 * #119950 * #119881 * #119229 * #119104 There...

wip: turn on native funcol by default to see which tests fail

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #119951 * #120118 * #120070 * #120043 * #120042 * #120025 * #119982 * #119950 * #119881 * #119229 * #119104 cc...

[dynamo] support rewriting dist.all_reduce with explicitly specified reduce op

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #120560 * __->__ #120181 * #120118 * #120370 Resolves 1 in https://github.com/pytorch/pytorch/issues/120082. cc @mrshenli @pritamdamania87 @zhaojuanmao @satgera @rohan-varma @gqchen @aazzolini @osalpekar @jiayisuse @H-Huang...

dist.group.WORLD

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #120560 * #120181 * #120118 * #120370 cc @mrshenli @pritamdamania87 @zhaojuanmao @satgera @rohan-varma @gqchen @aazzolini @osalpekar @jiayisuse @H-Huang @kwen2501 @awgu @penguinwu @fegin...

Remove Inductor IRs for legacy functional collectives

7

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #124992 This PR completely removes the Inductor IR for legacy functional collectives: - Removed the `CollectiveKernel` hiearchy and `Wait`, as well as...

### 🚀 The feature Leverage local disk for async snapshot. ### Motivation, pitch TorchSnapshot supports async snapshot, which allows training to resume before the storage I/O of a snapshot completes....