[API Compatibility] enhance paddle.enable_compat -part#79391
[API Compatibility] enhance paddle.enable_compat -part#79391Manfredss wants to merge 14 commits into
Conversation
…alls resolve to the paddle.compat.* implementations at runtime. assisted by Claude
risemeup1111
left a comment
risemeup1111
left a comment
There was a problem hiding this comment.
发现了需要修复后再合入的兼容性回归，具体问题已放在行级评论中：主要是 paddle.compat 的 lazy attribute 行为被破坏，以及已有的 compat public API 被删除。CI 当前仍有部分任务进行中，且已有 Windows-GPU build/test 失败，请后续一并确认。
| def __getattr__(name): | ||
| if name == "paddle_triton": | ||
| return paddle_triton_fun() |
There was a problem hiding this comment.
这里删掉了 distributions 的 lazy import 和最终的 AttributeError，导致两个回归：paddle.compat.distributions 在未预先 import paddle.compat.distributions 时会通过 __getattr__ 直接返回 None，未知属性也不再抛 AttributeError，会破坏 Python 模块属性协议以及已有的 paddle.compat.distributions 访问方式。请保留 distributions 分支，并让未知属性继续抛 AttributeError。
| def __getattr__(name): | |
| if name == "paddle_triton": | |
| return paddle_triton_fun() | |
| def __getattr__(name): | |
| if name == "paddle_triton": | |
| return paddle_triton_fun() | |
| if name == "distributions": | |
| import importlib | |
| module = importlib.import_module("paddle.compat.distributions") | |
| globals()[name] = module | |
| return module | |
| raise AttributeError(f"module {__name__!r} has no attribute {name!r}") |
| 'BatchNorm1d', | ||
| 'BatchNorm2d', | ||
| 'BatchNorm3d', | ||
| 'MultiheadAttention', |
There was a problem hiding this comment.
这里把 paddle.compat.nn.BatchNorm1D/2D/3D、小写别名以及 SmoothL1Loss 从 public compat surface 删除了，但这些 API 已经有现有调用和测试覆盖：test/legacy_test/test_api_compatibility_part1.py 在类定义时直接绑定 paddle.compat.nn.BatchNorm*，test/legacy_test/test_compat_smooth_l1_loss.py 也直接实例化 paddle.compat.nn.SmoothL1Loss。删除后这些测试会在 import/执行阶段变成 AttributeError，也会破坏已有用户代码。请恢复这些符号及其实现；如果目标是只做命名空间 alias，仍需要保留 compat 侧的实现作为 paddle.* alias 的目标。
建议形状：
__all__ = [
# ...
'BatchNorm1D', 'BatchNorm2D', 'BatchNorm3D',
'BatchNorm1d', 'BatchNorm2d', 'BatchNorm3d',
'SmoothL1Loss',
]
class BatchNorm1D(nn.BatchNorm1D):
... # 保留 torch-style eps/momentum/track_running_stats 适配
BatchNorm1d = BatchNorm1D
# 同步恢复 2D/3D 以及 SmoothL1Loss| return x | ||
|
|
||
| return paddle.nn.functional.unfold( | ||
| return _native(paddle.nn.functional, "unfold")( |
There was a problem hiding this comment.
这里删除 paddle.compat.nn.functional.smooth_l1_loss 会直接破坏已有 compat functional API：test/legacy_test/test_compat_smooth_l1_loss.py 覆盖了 F_compat.smooth_l1_loss(..., beta=...)、target=、size_average/reduce、beta=0 和 forbidden keywords。删除后这些用例和用户迁移代码都会变成 AttributeError，并且 paddle.enable_compat() 也无法把 paddle.nn.functional.smooth_l1_loss alias 到 torch-style 语义。请恢复该函数、__all__ 条目，以及所需的 warnings 和 _ReduceMode。
建议形状：
import warnings
if TYPE_CHECKING:
_ReduceMode: TypeAlias = Literal["mean", "sum", "none"]
__all__ = [
# ...
'unfold',
'smooth_l1_loss',
]
@ForbidKeywordsDecorator(
illegal_keys={"label", "delta", "is_huber", "name"},
func_name="paddle.compat.nn.functional.smooth_l1_loss",
correct_name="paddle.nn.functional.smooth_l1_loss",
)
def smooth_l1_loss(input, target, size_average=None, reduce=None, reduction='mean', beta=1.0):
# 保留原来的 torch-style beta/size_average/reduce 适配逻辑
...
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
CI 分析结果本轮达到时间上限，未完成全部失败原因分析；未分析完的 job 标记为“分析省略/待后续深挖”。 |
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
…mpat regression test
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
liuhao2638
left a comment
liuhao2638
left a comment
There was a problem hiding this comment.
已复查当前提交。此前指出的 paddle.compat lazy attribute、compat.nn BatchNorm/SmoothL1Loss、compat.nn.functional.smooth_l1_loss 删除问题在当前代码中已经恢复。
不过这轮发现了一个新的全局 alias 下的 P1 回归，具体见行级评论：smooth_l1_loss wrapper 需要通过 native 实现回调，避免 enable_compat() 后自引用到 compat wrapper。CI 目前仍有任务运行中，后续也请一并确认。
| for attr_name in getattr(compat_module, PUBLIC_ATTR_DECLARATION): | ||
| if attr_name.startswith("_"): | ||
| continue | ||
| compat_attr = getattr(compat_module, attr_name) |
There was a problem hiding this comment.
这里会把每个 paddle.compat.*.__all__ 符号 alias 到真实 paddle.* namespace，因此全局 paddle.enable_compat() 后 paddle.nn.functional.smooth_l1_loss 也会变成 compat wrapper。当前 wrapper 在函数尾部仍直接调用 paddle.nn.functional.smooth_l1_loss(..., delta=beta, is_huber=False)；alias 生效后这个名字已经指回 wrapper 自身，ForbidKeywordsDecorator 会拒绝 delta/is_huber（或进入自调用路径），导致迁移代码里的 paddle.nn.functional.smooth_l1_loss(x, y, beta=...) 失败。
请像 unfold/SDPA 一样通过 _native 调回 Paddle 原生实现，并补一个 enable_compat() 下的回归用例覆盖这个别名路径。建议形状：
if beta == 0:
return _native(paddle.nn.functional, "l1_loss")(
input, target, reduction=reduction
)
return _native(paddle.nn.functional, "smooth_l1_loss")(
input, target, reduction=reduction, delta=beta, is_huber=False
)
zhwesky2010
changed the title
|
/re-run all-failed |
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #79391 +/- ##
==========================================
Coverage ? 93.92%
==========================================
Files ? 6
Lines ? 181
Branches ? 0
==========================================
Hits ? 170
Misses ? 11
Partials ? 0 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
|
|
||
|
|
||
| @cache | ||
| def _register_compat_override(): |
There was a problem hiding this comment.
如果 paddle.compat.* 已经patch到 paddle.* 上了，这里还需要去额外操作吗
| # disable_compat() on exit would remove the finder and destroy the | ||
| # surrounding scoped enable (the runtime entry point of the migration | ||
| # workflow). So restore the finder to exactly how it was found. | ||
| if already_has_torch_proxy: |
| public_attrs = getattr(module, PUBLIC_ATTR_DECLARATION) | ||
| torch_module_name = module_info.name.replace( | ||
| PADDLE_PREFIX, TORCH_PREFIX, 1 | ||
| # Use the root-inclusive iterator: pkgutil.walk_packages skips the |
There was a problem hiding this comment.
这些注释清理一下吧，如果确需要注释的话，也只是简短意赅的关键1~2行，不用长篇大论的写
| _restore_paddle_namespace_aliases() | ||
|
|
||
|
|
||
| def enable_compat( |
There was a problem hiding this comment.
直接改这个可能有点不兼容，加个level参数吧
背景这个主要是解决当前的paddle框架中无法完全对齐的问题，当前Paddle API的对齐状态分为以下三种：
用户在编写paddle时，很难知道哪些是对齐的1，哪些是未对齐的2/3，这使得编写paddle代码成本仍较高。 开发思路目前考虑在enable_compat最后面加一个level参数，默认为1，对于：
针对 第三方库+存量Paddle代码 的用法：level=1，部分API无法对齐，需手动检查调整，主要为避免存量paddle代码出现不兼容 针对 第三方库+新增Paddle代码 的用法：level=2，完全对齐，统一按torch的形态即可 @SigureMo 看下这样的设计思路怎么样，有没有要调整的地方？比如 不加level直接按2来、加level但默认为2、新增一个其他API来作为这个开关不用放到enable_compat里？ |
SigureMo
left a comment
SigureMo
left a comment
There was a problem hiding this comment.
针对 第三方库+存量Paddle代码 的用法：level=1，部分API无法对齐，需手动检查调整，主要为避免存量paddle代码出现不兼容
针对 第三方库+新增Paddle代码 的用法：level=2，完全对齐，统一按torch的形态即可
我比较赞同这里的 level=1/2 的方案，主要是考虑到现存代码可能有部分已经使用 Paddle 原生 API 改写过了，此时一旦自动直接将这些 API 迁移到 paddle.compat API 实现可能会出现一些因为兼容性导致的非预期的问题，直接按 2 来我倒是有些担忧的，不过倒是后面可以考虑把 2 切默认（理想情况）
不过值得注意的是，paddle.enable_compat 这个 API 真正的挑战并不在于实现这些映射功能，而在于当环境中真的安装有 PyTorch 时，如何避免发生冲突（CI 中几乎测不到，只有几个使用 fake torch 模拟的 case），这是真实业务场景中会使用的，因此在实现时候需要特别注意下
| return x | ||
|
|
||
| return paddle.nn.functional.unfold( | ||
| return _native(paddle.nn.functional, "unfold")( |
There was a problem hiding this comment.
我有个疑问，既然是 patch，那么一旦开启 compat，所有相关 API 都会变吧？为什么当前改动仅限于 paddle.compat 下？非 compat 下的 API 理论上也会受到影响吧？（只是问题并不是像递归这么明显罢了）
当然不是让你把所有 API 改一遍，而是我在想这样做的可行性
There was a problem hiding this comment.
其实这也很好做，改造一下：
from typing import Callable, TypeVar
from typing_extensions import ParamSpec
from contextlib import contextmanager
T1 = TypeVar("T1")
T2 = TypeVar("T2")
P1 = ParamSpec("P1")
P2 = ParamSpec("P2")
COMPAT_API_ENABLED = False
def is_compat_api_enabled():
return COMPAT_API_ENABLED
@contextmanager
def compat_api_guard(enable: bool = True):
global COMPAT_API_ENABLED
if enable == COMPAT_API_ENABLED:
yield
return
original_value = COMPAT_API_ENABLED
COMPAT_API_ENABLED = enable
try:
yield
finally:
COMPAT_API_ENABLED = original_value
# compat/registry.py
def dispatch_compat_api(
compat_fn: Callable[P1, T1],
) -> Callable[[Callable[P2, T2]], Callable[P2, T2]]:
def native_fn_decorator(native_fn: Callable[P2, T2]) -> Callable[P2, T2]:
def maybe_use_compat_api(*args: P2.args, **kwargs: P2.kwargs) -> T2:
if is_compat_api_enabled():
# 关键在这里，灵活关掉 compat
return compat_api_guard(enable=False)(compat_fn)(*args, **kwargs) # type: ignore
return native_fn(*args, **kwargs)
return maybe_use_compat_api
return native_fn_decorator
def unfold_compat(x, y):
return unfold(x, y) + 1
@dispatch_compat_api(unfold_compat)
def unfold(x, y):
return x * y
with compat_api_guard():
print(unfold(1, 2)) # 3
with compat_api_guard(enable=False):
print(unfold(1, 2)) # 2随便搞的原型，仅供参考
| paddle.enable_compat() | ||
| try: | ||
| self.assertEqual(paddle.min(t).item(), 1.0) | ||
| self.assertEqual(paddle.max(t).item(), 3.0) | ||
| # dim form returns a namedtuple (values, indices) | ||
| r = paddle.min(t, dim=0) | ||
| self.assertEqual(r.values.item(), 1.0) | ||
| self.assertEqual(r.indices.item(), 1) | ||
| finally: | ||
| paddle.disable_compat() |
There was a problem hiding this comment.
这种单测不能直接用 @use_compat_guard 装饰器么？
There was a problem hiding this comment.
@Manfredss 对于整个框架内部，调用到这些compat系列API的地方（约30个），确实应该都是paddle naive版的，而这一系列API对外呈现的则是torch版的。
这个主要影响到的是 单测 + 部分调用paddle API的组合实现API。
可能得加大一下测试面，比如level=2下，paddle其他的API还能否跑过，除了paddle自身测试外，建议在paconvert里也进行测试（全局设置为level=2），测量全量API的行为是否正常。
…d the compat impls) get native, only external user code sees compat -- keeps paddle's own layers working
|
paddle.enable_compat 增加 level 参数： 背景：Paddle API 对齐分三类:
设计：
level=2 关键实现
测试
|
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
…Co-Authored-By: Claude <noreply@anthropic.com>
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
…red-by: Claude <noreply@anthropic.com>
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
|
/re-run all-failed |
There was a problem hiding this comment.
Pull request overview
This PR extends paddle.enable_compat() with a new compatibility “level=2” mode that (when enabled globally) aliases public paddle.compat.* APIs onto the live paddle / paddle.nn / paddle.nn.functional namespaces, using caller-aware dispatch to keep Paddle internal calls on native implementations. It also adds guards to avoid recursion when compat wrappers call back into aliased paddle.* APIs, and introduces comprehensive tests for aliasing + lifecycle behavior.
Changes:
- Add a level-based compat mode where global
enable_compat(level=2)aliasespaddle.*topaddle.compat.*using caller-aware dispatchers/proxies (and restores exactly on disable). - Add
compat_api_guard(enable=False)usage in compat wrappers to ensure internal calls hit native implementations under aliasing. - Add new test coverage (including fake modules) for namespace aliasing, scoped/global lifecycle behavior, and torch root API mapping.
Reviewed changes
Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| test/compat/test_compat_namespace_aliased.py | New end-to-end tests for level=2 aliasing, restoration, lifecycle correctness, and caller-aware dispatch behavior. |
| test/compat/fake_modules/torch_proxy_root_api_module.py | Fake module used to validate scoped imports + torch root API capture behavior. |
| python/paddle/compat/proxy.py | Implements level=2 namespace aliasing, caller-aware dispatch, compat suspension guard, root-package override coverage, and improved guard restoration logic. |
| python/paddle/compat/nn/transformer.py | Applies compat_api_guard(enable=False) to internal paths that must remain native under aliasing. |
| python/paddle/compat/nn/functional/sdpa.py | Guards compat SDPA implementation so its internals consistently use native implementations. |
| python/paddle/compat/nn/functional/init.py | Guards compat unfold to ensure internal calls remain native under level=2 aliasing. |
| python/paddle/compat/nn/init.py | Guards compat AvgPool* forward; adjusts Softmax.extra_repr (but currently introduces a bug). |
| python/paddle/compat/init.py | Exposes compat dispatch/guard helpers and guards several top-level compat functions to prevent recursion under aliasing. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| def extra_repr(self) -> str: | ||
| return f"dim={self._dim}" | ||
| return f"dim={self.dim}" | ||
|
|
| # (live module, attr name) -> original value (or ``_MISSING``) for every | ||
| # ``paddle.*`` symbol aliased to its ``paddle.compat.*`` counterpart, so the | ||
| # paddle namespace can be restored exactly on ``disable_compat()``. | ||
| _PADDLE_NAMESPACE_SAVED: dict[tuple[types.ModuleType, str], Any] = {} | ||
|
|
|
/re-run all-failed |
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
|
/re-run all-failed |
1 similar comment
|
/re-run all-failed |
也就是确保paddle.enable_compat(level=2)仅修改对外的 |
liuhao2638
left a comment
liuhao2638
left a comment
There was a problem hiding this comment.
已复查当前提交。此前由 risemeup1111 提出的 paddle.compat lazy attribute 和 compat public API 删除问题，在当前代码中看起来已恢复；本轮新增的阻塞问题已放在行级评论里，主要是公开 API 装饰器导致 API spec 签名变化并触发 Static-Check 失败。
非行级：测试覆盖缺口不能准确附着到单一 diff 行。当前 Paddle 侧新增测试覆盖了几个代表性层和 Tensor 方法，但还没有把 enable_compat(level=2) 写入约 10 个会在内部调用 split/max/min 的组合 API 用例（例如 vsplit/hsplit/dsplit 及相关组合路径），也没有在本 PR 可见范围内给出 PaConvert 全量 API 在 level=2 下通过的结果。请补齐这些回归用例或在 PR 中给出对应测试记录，确保 level=2 只影响外部 paddle.* 调用，不波及 Paddle 内部组合 API。
非行级：PR 描述不在 changed diff line 上。当前描述仍写 enable_compat() 后 paddle.X 即变为 compat、通过 _native helper 回调 native、以及 flash_attention.py 显式传 F.softmax(product, -1)；但当前实现实际是默认 level=1、只有 enable_compat(level=2) 才启用 paddle.* namespace alias，并通过 compat_api_guard/caller-aware dispatch 处理内部调用，flash_attention.py 当前也仍是 F.softmax(product)。建议同步描述，避免后续 reviewer 和用户按旧语义理解这个开关。
| @contextmanager | ||
| def compat_api_guard(enable: bool = True) -> Generator[None, None, None]: | ||
| """Suspend/restore compat dispatch for the current thread / async task only | ||
| (task-local). Compat APIs use ``@compat_api_guard(enable=False)`` so their own | ||
| ``paddle.*`` calls hit native instead of re-dispatching into compat.""" | ||
| token = _COMPAT_SUSPENDED.set(not enable) | ||
| try: | ||
| yield | ||
| finally: | ||
| _COMPAT_SUSPENDED.reset(token) |
There was a problem hiding this comment.
compat_api_guard 现在由 contextlib.contextmanager 生成后直接当公开 API 装饰器使用，会把被装饰函数在 inspect.getfullargspec() 下暴露成 *args/**kwds。当前 Static-Check 已经因此在 paddle.compat.nn.Unfold.forward 和 paddle.compat.nn.functional.scaled_dot_product_attention 上报 api_params_diff，这会阻塞 API 兼容检查。请改成同时支持 with 和装饰器、并保留原函数签名的实现，或把这些公开 API 内的 guard 改成函数体里的 with。
一个集中修复的形状如下：
| @contextmanager | |
| def compat_api_guard(enable: bool = True) -> Generator[None, None, None]: | |
| """Suspend/restore compat dispatch for the current thread / async task only | |
| (task-local). Compat APIs use ``@compat_api_guard(enable=False)`` so their own | |
| ``paddle.*`` calls hit native instead of re-dispatching into compat.""" | |
| token = _COMPAT_SUSPENDED.set(not enable) | |
| try: | |
| yield | |
| finally: | |
| _COMPAT_SUSPENDED.reset(token) | |
| class _CompatApiGuard: | |
| def __init__(self, enable: bool = True) -> None: | |
| self.enable = enable | |
| self._token: Any | None = None | |
| def __enter__(self) -> None: | |
| self._token = _COMPAT_SUSPENDED.set(not self.enable) | |
| def __exit__(self, exc_type: Any, exc: Any, tb: Any) -> None: | |
| token = self._token | |
| self._token = None | |
| if token is not None: | |
| _COMPAT_SUSPENDED.reset(token) | |
| def __call__(self, func: Any) -> Any: | |
| @wraps(func) | |
| def wrapper(*args: Any, **kwargs: Any) -> Any: | |
| with type(self)(self.enable): | |
| return func(*args, **kwargs) | |
| wrapper.__signature__ = inspect.signature(func) | |
| return wrapper | |
| def compat_api_guard(enable: bool = True) -> _CompatApiGuard: | |
| """Suspend/restore compat dispatch for the current thread / async task only | |
| (task-local). Compat APIs use ``@compat_api_guard(enable=False)`` so their own | |
| ``paddle.*`` calls hit native instead of re-dispatching into compat.""" | |
| return _CompatApiGuard(enable) |
…omposite APIs enable_compat(level=2) only aliases the outward paddle.* surface; composite APIs (vsplit/chunk/quantile/nan_to_num/nll_loss/embedding/histogram*/GRUCell/LSTMCell/ clip_grad_norm_) that internally call split/sort/max/equal/unique must stay native via caller-aware dispatch. Verifies zhwesky2010's review request. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
|
/re-run all-failed |
1 similar comment
|
/re-run all-failed |
…ghtweight core The newly-added test hit the added-UT CI gate (--repeat-until-fail 3 --timeout 15/20) on Linux-CPU/XPU: heavy/exotic ops (GRUCell/LSTMCell with unseeded randn, histogramdd, in-place embedding renorm) caused nondeterministic abort+timeout, while Windows/Mac (no such gate) passed. Reduce to fixed-input, lightweight core composites (vsplit/hsplit/dsplit/chunk, quantile, nan_to_num, histogram_bin_edges, nll_loss) that still verify internal split/sort/min/max/equal stay native. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
risemeup1111
left a comment
risemeup1111
left a comment
There was a problem hiding this comment.
已复查新提交。Paddle 侧新增了 test_compat_level2_internal_composites.py，覆盖 vsplit/hsplit/dsplit/chunk、quantile、nan_to_num、histogram_bin_edges、nll_loss 等在 level=2 下仍需走 native 内部调用的轻量路径，上一轮关于 Paddle 单测覆盖的缺口已部分补齐。本轮没有新增行级评论；此前公开 API 签名变化的问题仍由已有行级评论覆盖。
非行级：跨仓 PaConvert 验证不在本 PR diff 行内。关联的 PaConvert PR 已改为注入 paddle.enable_compat(level=2)，但当前仍未看到全量 API/CI 通过结果，且关联 PR 的 Common API Consistency、Distributed、Modeltest 等检查仍处于失败/未完成状态。请补充通过记录或修复对应失败后再把这里作为完整验证闭环。
非行级：PR 描述不在 changed diff line 上。当前描述仍按 plain paddle.enable_compat() 即启用 namespace alias、_native helper、以及 flash_attention.py 显式 F.softmax(product, -1) 来表述；当前实现实际是默认 level=1、level=2 才启用 alias，并通过 compat_api_guard/caller-aware dispatch 处理内部调用。建议同步描述到当前实现，避免 reviewer 和用户按旧语义理解。
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
|
/re-run all-failed |
1 similar comment
|
/re-run all-failed |
risemeup1111
left a comment
risemeup1111
left a comment
There was a problem hiding this comment.
已复查本次测试更新：test_compat_level2_internal_composites.py 补充了 tensor_split 覆盖，Paddle 侧内部组合 API 的轻量回归覆盖进一步收窄；未发现这次测试改动带来新的行级问题。此前最早提出的 paddle.compat lazy attribute 和 compat public API 删除问题在当前代码中仍为已恢复。公开 API 签名问题当前实现未变化，仍由已有行级讨论覆盖，本轮不重复评论。
非行级：跨仓 PaConvert 验证不在本 PR diff 行内。关联 PaConvert PR 目前已继续调整 enable_compat(level=2) 注入/重置逻辑，但当前 CI Unittest、Common API Consistency Check、Modeltest Check 仍失败，Distributed/GPU 也尚未全部完成。请补充全量 API/CI 通过记录，或先修复关联失败后再把这里作为完整验证闭环。
非行级：PR 描述不在 changed diff line 上。当前描述仍按 plain paddle.enable_compat() 即启用 namespace alias、_native helper、以及 flash_attention.py 显式 F.softmax(product, -1) 来表述；当前实现实际是默认 level=1、level=2 才启用 alias，并通过 compat_api_guard/caller-aware dispatch 处理内部调用，flash_attention.py 当前也仍是 F.softmax(product)。建议同步描述到当前实现，避免 reviewer 和用户按旧语义理解。
CI报告基于以下代码生成（30分钟更新一次）: 1 Required任务 : 45/48 通过
2 失败详情🔴 Static-Check / Test — PR问题（置信度: 高）分析器: 通用分析(fallback)
关键日志:
修复建议:
关联变更: 🔴 Fleet Unit test (multi-card) — 环境问题（置信度: 高）分析器: 通用分析(fallback)
关键日志:
修复建议:
关联变更: 未发现与本 PR 修改文件的直接关联。 🔴 Fleet Unit test (single card) — 环境问题（置信度: 高）分析器: 通用分析(fallback)
关键日志:
修复建议:
关联变更: 未发现与本 PR 修改文件的直接关联。 |
|
/re-run all-failed |
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
|
/re-run all-failed |
1 similar comment
|
/re-run all-failed |
risemeup1111
left a comment
risemeup1111
left a comment
There was a problem hiding this comment.
已复查最新提交。compat_api_guard 已从 contextmanager 改为同时支持上下文和装饰器的 class，并在装饰器路径保留 __signature__；我用等价最小复现确认 inspect.getfullargspec 不再退化成 *args/**kwargs，此前公开 API 签名变化的 P1 问题看起来已由新 commit 修复。最早提出的 paddle.compat lazy attribute 和 compat public API 删除问题当前也仍为已恢复。本轮没有新增行级评论。
非行级：跨仓 PaConvert 验证不在本 PR diff 行内。当前关联 PaConvert PR 的 checks 已显示全部通过，我倾向于认为验证证据已补齐；但该项此前作为 P2 跟进意见提出，仍需要作者在本 PR 中回复确认对应验证记录。
处理要求：请针对该评论进行回复（同意并已修改请回复 Done，不同意请说明理由）。
非行级：PR 描述不在 changed diff line 上。当前描述仍按 plain paddle.enable_compat() 即启用 namespace alias、_native helper、以及 flash_attention.py 显式 F.softmax(product, -1) 来表述；当前实现实际是默认 level=1、level=2 才启用 alias，并通过 compat_api_guard/caller-aware dispatch 处理内部调用，flash_attention.py 当前也仍是 F.softmax(product)。请同步描述到当前实现。
处理要求：请针对该评论进行回复（同意并已修改请回复 Done，不同意请说明理由）。
PaddlePaddle-bot
left a comment
PaddlePaddle-bot
left a comment
There was a problem hiding this comment.
🤖 Paddle-CI-Agent | pr_review |
2026-07-08 19:48:21
📋 Review 摘要
PR 概述：为 paddle.enable_compat(level=2) 增加 paddle 命名空间到 paddle.compat 的 caller-aware alias，并补充内部 composite 调用保护与测试。
变更范围：python/paddle/compat、paddle.compat.nn/functional、compat namespace 回归测试。
影响面 Tag：User Experience
问题
未发现阻塞性问题。PR 规范问题在下面章节报,不要在这里重复
历史 Findings 修复情况
| Finding | 问题 | 状态 |
|---|---|---|
| F1 | enable_compat() 默认仍是 level=1，与 plain paddle.enable_compat() 迁移入口不一致。 |
|
| F2 | dispatcher 使用 @wraps(native_fn)，level=2 后公开 paddle.* 仍暴露 native 元数据/签名。 |
|
| F3 | 新测试在模块导入阶段永久追加 fake_modules 到全局 sys.path。 |
📝 PR 规范检查
标题 Tag API Compatibility 不在当前 Paddle 模板的 PR Category / PR Types 枚举中。描述包含必填四段，且“是否引起精度变化”为“否”，结构符合模板。
标题建议（可直接复制）：
[User Experience] enhance paddle.enable_compat level=2 namespace aliasing
总体评价
本轮按 Python 公开 API/兼容性路径检查了 namespace alias、内部 guard、scope lifecycle、Tensor method patch 和新增测试。未新增 inline findings；历史 3 条仍未在当前 diff 中修复，建议后续一并收敛。
|
/re-run all-failed |
PR Category
User Experience
PR Types
Improvements
Description
paddle.enable_compat()currently only installs asys.meta_pathfinder so thatimport torchresolves to PaddlePaddle (a torch→paddle proxy); it does not touchthe real
paddle.*namespace, sopaddle.sort/paddle.min/... stay paddle-nativeeven while compat is enabled.
Because
paddle.Xandpaddle.compat.Xshare identical positional parameter types, acall site cannot be auto-classified as "native intent" vs "compat intent", so the clean
design is a global switch: after
enable_compat(),paddle.Xispaddle.compat.Xand its arguments are interpreted as the torch-aligned compat arguments. Users write
PyTorch-style code and no longer need to care whether an API is
paddle.*orpaddle.compat.*. (This also backs the PaConvert migration flow: prefix-onlytorch.X → paddle.X+ an injectedpaddle.enable_compat().)What this PR does
compat/proxy.py): onenable_compat()(global scope only),alias every public
paddle.compat.*symbol onto the livepaddle/paddle.nn/paddle.nn.functionalmodules viasetattr, mirroring_register_compat_override(
torch.X → paddle.compat.X).disable_compat()perfectly restores the originalnamespace (including deleting symbols that did not exist before —
paddle.slogdet,paddle.nn.AvgPool1d/2d/3d,paddle.nn.MultiheadAttention)._nativehelper): compat wrappers that internally call backinto a name they alias (
paddle.compat.min→paddle.min,Unfold/transformerF.linear/F.softmax,unfold,scaled_dot_product_attention) now route through_native(module, name)so they always reach the genuine native impl — avoidinginfinite recursion /
ForbidKeywordsrejection / a silent double-transpose, whilekeeping
paddle.X is paddle.compat.X._register_compat_override) now alsoprocess the
paddle.compatroot package thatpkgutil.walk_packagesskips, so thetop-level functions (sort/min/max/split/median/unique/...) are mapped for both global
and scoped enable.
enable_compat(scope=...)while already global no longer silently downgrades to scoped;
use_compat_guardfaithfully restores the original finder state (so importing a scoped module no longer
destroys the surrounding scoped enable); and a leaked torch
ProxyModuleis clearedwhen restoring to scoped state.
nn/functional/flash_attention.py): aliasingpaddle.nn.functional.softmaxto the torch-style compat softmax (whosedim=Nonedefault picks a different axis than native
axis=-1) silently corrupted the sdpamath backend, because
_math_attentioncalledF.softmax(product)with no axis.Pass the axis explicitly (
F.softmax(product, -1)) so attention stays correct underenable_compat(); with compat off this is identical to the previous behavior.是否引起精度变化
否