Description

Summary: Multiple dataframe pipe tests failing due to numpy binary incompatibility on CPU and sGPU platforms.

Test Class: test/test_datapipe.py::TestDataFramesPipes
Number of Failing Tests: 6
Platform: CPU, sGPU (CUDA)
Test Type: Unit Test

Version Information:

• PyTorch Commit: 6bdd8c9
• Branch: main
• Test Date: 2026-01-14
• Sprint: Sprint 24

Failure Pattern:
All tests failing with the same error - single root cause

Common Error:
code
ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject
This exception is thrown by _iter_ of ExampleAggregateAsDataFrames(...)
code

Failing Tests:
1. test_batch
2. test_capture
3. test_collate
4. test_filter
5. test_shuffle
6. test_unbatch

Steps to Reproduce:
code
TEST_CONFIG=cpu python3 test/run_test.py -i test_datapipe
TEST_CONFIG=cuda python3 test/run_test.py -i test_datapipe
code

Expected Result:
All tests should pass with compatible numpy/pandas versions

Actual Result:
All tests fail with numpy binary incompatibility error

Root Cause Analysis:
This error indicates a version mismatch between numpy used to compile pandas and runtime numpy. Same issue as TestCaptureDataFrame - numpy/pandas binary incompatibility.

Potential Solutions:
1. Reinstall pandas compiled against the current numpy version
2. Downgrade numpy to version compatible with current pandas build
3. Use pip install --no-build-isolation when installing pandas

Priority: P3