PaddlePaddle · Echo-Nie · Aug 12, 2026
diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-78932/README.md b/swe-paddle/tasks/PaddlePaddle__Paddle-78932/README.md
@@ -0,0 +1,44 @@
+# PaddlePaddle__Paddle-78932
+
+This directory converts Paddle PR #78932 into a SWE-Paddle community task candidate.
+
+## Source
+
+| Field | Value |
+| --- | --- |
+| Repo | `PaddlePaddle/Paddle` |
+| PR | [78932](https://github.com/PaddlePaddle/Paddle/pull/78932) |
+| PR title | `[API Compatibility] Support vararg and add alias for paddle.io.TensorDataset` |
+| Base commit | `7b7e53fd28956700e5ed1ce68eb2aaeb59829777` |
+| Merged at | `2026-05-11T08:51:45Z` |
+| Task type | `feature_enhancement` |
+| Resource | CPU |
+
+## Summary
+
+Allow `TensorDataset` to accept one or more tensors directly and expose it through `paddle.utils.data`, while preserving the existing list/tuple calling convention.
+
+## Why This Is A Good SWE-Paddle Candidate
+
+- The task reflects a common data-loading compatibility issue with clear inputs and observable dataset behavior.
+- The change must distinguish a single Tensor from a list/tuple and from multiple positional tensors without breaking existing calls.
+- The source PR provides real tests for list-based construction, single- and multi-Tensor varargs, item structure, dataset length, and public alias availability.
+- The tests run deterministically on CPU without workers, external datasets, network access, or distributed devices.
+
+## Files
+
+- `proposal.md`: candidate proposal for maintainer triage.
+- `instruction.md`: self-contained problem statement for the coding agent.
+- `solution/code.patch`: gold patch from the merged PR.
+- `tests/test.patch`: exact upstream test changes exposing the target behavior.
+- `tests/test.sh`: minimal target test command.
+- `environment/README.md`: environment notes for reproduction.
+- `README.md`: task overview and verification entrypoint.
+
+## Verification
+
+```bash
+bash tests/test.sh
+```
+
+Expected behavior: applying `tests/test.patch` to `base_commit` should preserve the existing list-based cases but fail on direct Tensor arguments and the missing public alias; applying both `tests/test.patch` and `solution/code.patch` should pass all target tests.
diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-78932/environment/README.md b/swe-paddle/tasks/PaddlePaddle__Paddle-78932/environment/README.md
@@ -0,0 +1,27 @@
+# Environment Notes
+
+This candidate is part of the SWE-Paddle community task set.
+
+## Expected Environment
+
+- Repository: `PaddlePaddle/Paddle`
+- Base commit: `7b7e53fd28956700e5ed1ce68eb2aaeb59829777`
+- Resource: CPU
+- GPU required: no
+- Build path: use an installed CPU Paddle wheel for the native runtime and load the target Python behavior from the source checkout; a Paddle source rebuild is not required.
+
+## Run Order
+
+1. Check out `PaddlePaddle/Paddle` at the base commit.
+2. Apply `tests/test.patch`.
+3. Run `bash tests/test.sh`; the direct Tensor and public alias cases should fail before the fix.
+4. Apply `solution/code.patch`.
+5. Run `bash tests/test.sh` again; all target tests should pass after the gold patch.
+
+## Minimal Test Command
+
+```bash
+bash tests/test.sh
+```
+
+The verifier is responsible for deriving stable F2P and P2P node IDs from repeated runs.
diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-78932/instruction.md b/swe-paddle/tasks/PaddlePaddle__Paddle-78932/instruction.md
@@ -0,0 +1,19 @@
+# 让 TensorDataset 支持常见的多参数写法
+
+## 详细描述
+
+目前 `paddle.io.TensorDataset` 需要把所有 Tensor 先放进一个 list 或 tuple 再传入。很多数据加载代码会直接使用 `TensorDataset(features, labels)` 这种写法，在 Paddle 中会报参数错误。只传一个 Tensor 时问题更隐蔽：Tensor 会被当成一组 Tensor 来处理，导致数据集长度和取出的数据结构不正确。
+
+同时，使用 `paddle.utils.data` 的代码目前也找不到 `TensorDataset`。需要补齐这些常见调用方式，且不影响 Paddle 现有的 list/tuple 写法。
+
+## 验收说明
+
+- 传入多个 Tensor 时，数据集长度应取各 Tensor 的第一维，按索引读取时应返回包含对应行的 tuple。
+- 直接传入单个 Tensor 时，应得到只包含一项的 tuple，数据集长度仍为该 Tensor 的第一维。
+- 原有 list/tuple 调用必须继续正常工作，并且 `paddle.utils.data.TensorDataset` 应可正常访问。
+
+## 技术要求
+
+- 熟悉 Python 位置参数、可变参数和 decorator。
+- 熟悉 Paddle `Dataset` 和 Tensor 索引行为。
+- 能够使用现有 DataLoader 与 API compatibility 测试验证新旧调用方式。
diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-78932/proposal.md b/swe-paddle/tasks/PaddlePaddle__Paddle-78932/proposal.md
@@ -0,0 +1,55 @@
+# Task Proposal: PaddlePaddle__Paddle-78932
+
+## 1. 来源信息
+
+- Instance ID：`PaddlePaddle__Paddle-78932`
+- PR 链接：https://github.com/PaddlePaddle/Paddle/pull/78932
+- PR 标题：`[API Compatibility] Support vararg and add alias for paddle.io.TensorDataset`
+- `base_commit`：`7b7e53fd28956700e5ed1ce68eb2aaeb59829777`
+- merged 时间：`2026-05-11T08:51:45Z`
+- 你的身份：熟悉该模块的 contributor
+- 后续联系人：TBD
+
+## 2. 问题一句话
+
+`TensorDataset` 无法直接接收一个或多个 Tensor 作为独立参数，并且不能从 `paddle.utils.data` 使用同名 API。
+
+## 3. 为什么适合作为 SWE-Paddle 样本
+
+- **真实性**：目标调用方式在数据加载代码迁移时很常见，当前会直接报参数错误或得到错误的数据集长度。
+- **代表性**：要求同时兼容 list/tuple、单 Tensor 和多 Tensor 三种入参形式，与常见的 Python API 兼容问题一致。
+- **边界清楚**：修改集中在 `TensorDataset` 的参数处理和公开导出，不涉及 DataLoader worker 或数据集外部资源。
+- **非平凡性**：单 Tensor 在 Python 中可迭代，不能简单当作 Tensor 列表处理；还要保证共享参数适配逻辑不破坏已有 API。
+- **环境友好性**：所有目标测试只使用小型 CPU Tensor，可稳定重复。
+
+## 4. 任务类型和标签
+
+- 任务类型：`feature_enhancement`
+- 执行后端：`cpu`
+- 设备范围：`cpu_only`
+- 模块标签：`[data, dataset, api-compatibility]`
+
+## 5. 验证思路
+
+- 目标测试命令：`bash tests/test.sh`
+- 目标测试文件：`test/ai_edited_test/test_ai_dataloader.py` 和 `test/legacy_test/test_paddle_utils_data.py`
+- 修复前预期：现有 list 调用正常；多 Tensor 位置参数报错，单 Tensor 被误解为可迭代的 Tensor 集合，`paddle.utils.data.TensorDataset` 不可用。
+- 修复后预期：旧 list 调用保持不变，单 Tensor 和多 Tensor 调用返回正确长度与 tuple item，新公开入口可正常访问。
+- P2P 候选：`test_tensor_dataset_basic`、`test_tensor_dataset_1d`、`test_tensor_dataset_iter`。
+- F2P 候选：`test_tensor_dataset_varargs`、`test_tensor_dataset_varargs_single`、`TestAlias::test_compatibility`。
+
+## 6. 环境与资源
+
+- 资源需求：CPU
+- Paddle 来源：`PaddlePaddle/Paddle` source checkout at `base_commit`
+- 是否能提供 Docker：暂无
+- patch 类型：Python-only
+- 最小测试命令：`bash tests/test.sh`
+- 是否有 oracle 日志：由 SWE-Paddle verifier 结果另行维护
+
+## 7. 风险自查
+
+- 泄露风险：任务描述只说明用户可见的入参方式和数据集行为，未说明 Gold patch 的具体实现。
+- 环境风险：历史 checkout 与已安装 wheel 可能有 Python 模块差异，cross verifier 需要使用受控的 source overlay。
+- flaky 风险：测试只校验 shape、长度、tuple 结构和 API 可访问性，不依赖随机数值、worker 时序或外部资源。
+- 拆分风险：varargs 和 `paddle.utils.data` 导出都围绕同一个 `TensorDataset` API 兼容问题，与来源 PR 的单一目标一致。
diff --git a/swe-paddle/tasks/PaddlePaddle__Paddle-78932/solution/code.patch b/swe-paddle/tasks/PaddlePaddle__Paddle-78932/solution/code.patch
@@ -0,0 +1,92 @@
+diff --git a/python/paddle/io/dataloader/dataset.py b/python/paddle/io/dataloader/dataset.py
+index e7519ae2cfe0401b029d2f91a904915a2894afa0..2a8ecb8c1bb0659ec90d46697ed05f9483a91d96 100755
+--- a/python/paddle/io/dataloader/dataset.py
++++ b/python/paddle/io/dataloader/dataset.py
+@@ -25,9 +25,10 @@ from typing import (
+     TypeVar,
+ )
+
+-from typing_extensions import Never, TypeVarTuple, Unpack
++from typing_extensions import Never, TypeVarTuple, Unpack, overload
+
+ import paddle
++from paddle.utils.decorator_utils import variadic_tensor_decorator
+
+ from ... import framework
+
+@@ -319,6 +320,13 @@ class TensorDataset(Dataset["Tensor"]):
+
+     tensors: Sequence[Tensor]
+
++    @overload
++    def __init__(self, tensors: Sequence[Tensor]) -> None: ...
++
++    @overload
++    def __init__(self, *tensors: Tensor) -> None: ...
++
++    @variadic_tensor_decorator('tensors', 1)
+     def __init__(self, tensors: Sequence[Tensor]) -> None:
+         if not framework.in_dynamic_mode():
+             raise RuntimeError(
+diff --git a/python/paddle/utils/data/__init__.py b/python/paddle/utils/data/__init__.py
+index 34502b82e20c8dc0235e3a842d10649824dbd2af..49358baf5bf1033accb2530f613049304dd8c810 100644
+--- a/python/paddle/utils/data/__init__.py
++++ b/python/paddle/utils/data/__init__.py
+@@ -22,6 +22,7 @@ from .dataset import (
+     Dataset,
+     IterableDataset,
+     Subset,
++    TensorDataset,
+     random_split,
+ )
+ from .sampler import (
+@@ -44,4 +45,5 @@ __all__ = [
+     'RandomSampler',
+     'Sampler',
+     'SequentialSampler',
++    'TensorDataset',
+ ]
+diff --git a/python/paddle/utils/data/dataset.py b/python/paddle/utils/data/dataset.py
+index f4a5f8a70c43162fa3886f33d187d0bca68eebe6..469280a61bb22f9601eec3b26670e9e0087d9787 100644
+--- a/python/paddle/utils/data/dataset.py
++++ b/python/paddle/utils/data/dataset.py
+@@ -18,5 +18,6 @@ from paddle.io import (
+     Dataset as Dataset,
+     IterableDataset as IterableDataset,
+     Subset as Subset,
++    TensorDataset as TensorDataset,
+     random_split as random_split,
+ )
+diff --git a/python/paddle/utils/decorator_utils.py b/python/paddle/utils/decorator_utils.py
+index b7f55e0cf05f66d7c02d073a76e0918803c3a6d7..bda544574049e16960c50709466ea0ff17ad8a55 100644
+--- a/python/paddle/utils/decorator_utils.py
++++ b/python/paddle/utils/decorator_utils.py
+@@ -1001,6 +1001,7 @@ def use_first_signature(
+
+ def variadic_tensor_decorator(
+     param_name: str,
++    param_pos: int = 0,
+ ) -> Callable[[Callable[_InputT, _RetT]], Callable[_InputT, _RetT]]:
+     """
+     Decorator to handle variadic tensor arguments.
+@@ -1018,14 +1019,14 @@ def variadic_tensor_decorator(
+         def wrapper(*args: _InputT.args, **kwargs: _InputT.kwargs) -> _RetT:
+             # PyTorch usage: variadic tensor arguments
+             if len(args) >= 1 and isinstance(
+-                args[0], (paddle.Tensor, paddle.pir.Value)
++                args[param_pos], (paddle.Tensor, paddle.pir.Value)
+             ):
+-                kwargs[param_name] = list(args)
+-                args = ()
++                kwargs[param_name] = list(args[param_pos:])
++                args = args[:param_pos]
+             # Paddle usage: list/tuple argument
+-            elif len(args) >= 1 and isinstance(args[0], (list, tuple)):
+-                kwargs[param_name] = args[0]
+-                args = ()
++            elif len(args) >= 1 and isinstance(args[param_pos], (list, tuple)):
++                kwargs[param_name] = args[param_pos]
++                args = args[:param_pos]
+             return func(*args, **kwargs)
+
+         wrapper.__signature__ = inspect.signature(func)