Tarek Ziade 37ed014053 Retire test_multi_gpu_data_parallel_forward (#48508) ...

`nn.DataParallel` cannot replicate or gather the forward patterns and
output types modern Transformers models use, so this common test measures
PyTorch's legacy DP path rather than model correctness.

DataParallel is a Trainer feature, not a per-model contract, and it keeps
its coverage there (tests/trainer/test_trainer.py). This removes the common
test and all 66 overrides; no `src/` behavior changes.

2026-09-04 11:57:36 +00:00

..

__init__.py

Move test model folders (#17034)

2022-05-03 14:42:02 +02:00

test_modeling_splinter.py

Retire test_multi_gpu_data_parallel_forward (#48508)

2026-09-04 11:57:36 +00:00

test_tokenization_splinter.py

rm slow tokenizers (#40936)

2025-11-27 19:24:50 +01:00