mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 12:54:11 +08:00
4661200125
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
37 lines
1.1 KiB
Python
37 lines
1.1 KiB
Python
"""
|
|
Configuration module for torch.export.export.
|
|
|
|
This module contains various configuration flags and settings that control torch.export's
|
|
behavior, including:
|
|
- Runtime behavior flags
|
|
- Debugging and development options
|
|
"""
|
|
|
|
import sys
|
|
from typing import Any, TYPE_CHECKING
|
|
|
|
from torch.utils._config_module import install_config_module
|
|
|
|
|
|
# this flag controls whether we use new functional tracer. It
|
|
# should be True in the long term.
|
|
use_new_tracer_experimental = False
|
|
|
|
# this flag is used to control whether we want to instrument
|
|
# fake tensor creation to track potential leaks. It is off
|
|
# by default, but user can turn it on to debug leaks.
|
|
detect_non_strict_fake_tensor_leaks = False
|
|
|
|
# error on potentially pre-dispatch/non-strict tracing limitation
|
|
# this type of error usually happens when we encounter an op
|
|
# that we don't know how to proxy, resulting in untracked fake tensors
|
|
error_on_lifted_constant_tensors = True
|
|
|
|
if TYPE_CHECKING:
|
|
from torch.utils._config_typing import * # noqa: F401, F403
|
|
|
|
def _make_closure_patcher(**changes: Any) -> Any: ...
|
|
|
|
|
|
install_config_module(sys.modules[__name__])
|