Description

Summary: DataLoader tests failing with worker process errors on CPU and sGPU platforms.

Test Class: test/test_dataloader.py::TestDataLoader
Number of Failing Tests: 2
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:
Tests failing with related errors - likely common root cause (numpy incompatibility in worker processes)

Common Error:
code
ValueError: Caught ValueError in DataLoader worker process 0.
Original Traceback (most recent call last):
File "/miniconda/envs/cuda_torch_build/lib/python3.12/site-packages/torch/utils/data/_utils/worker.py", line 358, in _worker_loop
data = fetcher.fetch(index)
File "/miniconda/envs/cuda_torch_build/lib/python3.12/site-packages/torch/utils/data/_utils/fetch.py", line 35, in fetch
code

Failing Tests:
1. test_multiprocessing_iterdatapipe
2. test_segfault

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

Expected Result:
Tests should pass without worker process errors

Actual Result:
Tests fail with ValueError in DataLoader worker process

Root Cause Analysis:
The DataLoader worker processes are encountering errors when trying to fetch data. This is likely related to the same numpy/pandas binary incompatibility issue affecting other datapipe tests.

Priority: P3