3 Commits

Author	SHA1	Message	Date
Tugsbayasgalan Manlaibaatar	4661200125	[RELAND v2] Close some sources of fake tensors (#164372) ... Changelog: 1. When we run into an operation we didn't proxy, we end up emitting fake constants. We error under a config and we disable the config for some internal users. The reason we want to error is this signals a coverage problem we need to address but at the same time, we don't wnat to be disruptive to already working flows. 2. Previous attribute mutation detection logic in non-strict didn't account for nested module structure. This fixes silent incorrectness issue of exporting esm and qwen in non-strict and some torchbench models like levit_128 and demucs. 3. Previous logic didn't work on the cases where we mutate a container attribute as the previous approach used to pytree over old and new attributes resulting in length mismatch. We gracefully handle this now. Differential Revision: [D83673054](https://our.internmc.facebook.com/intern/diff/D83673054) Pull Request resolved: https://github.com/pytorch/pytorch/pull/164372 Approved by: https://github.com/avikchaudhuri	2025-10-02 18:58:52 +00:00
Tugsbayasgalan Manlaibaatar	b756b580fb	Improve fake tensor leakage detection in export by not relying on gc too much (#163516) ... Previously we relied on gc to get the snapshot of fake tensors before and after export to get list of fake tensors that are created during export. This caused some flakiness in our test suite (https://github.com/pytorch/pytorch/issues/162232). it seems super hard to make gc deterministic, so we just instrument fake tensor creation which seems lot better. In addition, it is also quite faster than previous approach becuase we are no longer manually triggering garbage collector. Differential Revision: [D82966648](https://our.internmc.facebook.com/intern/diff/D82966648) Pull Request resolved: https://github.com/pytorch/pytorch/pull/163516 Approved by: https://github.com/ezyang	2025-09-22 22:04:24 +00:00
Tugsbayasgalan Manlaibaatar	0e9f9c3a61	Fix inconsistent test and add new tracer as config (#162558) ... It is better to have the new tracer as global config that can be manipulated easily. Also I believe dynamo-like config infra is useful instead of relying on custom way of patching stuff. Differential Revision: [D82478649](https://our.internmc.facebook.com/intern/diff/D82478649) Pull Request resolved: https://github.com/pytorch/pytorch/pull/162558 Approved by: https://github.com/zhxchen17 ghstack dependencies: #162557	2025-09-17 17:01:48 +00:00