From b9a2637e495c327caa9850eb4b418606496644e6 Mon Sep 17 00:00:00 2001 From: RunguoLi Date: Mon, 21 Sep 2026 16:14:14 -0500 Subject: [PATCH 1/2] [example] fix Pytorch_DDP/Pytorch_ZeRO plans in gpt gemini demo `train_gpt_demo.py` asserts `args.tp_degree == 1` in the Pytorch_* branch, but the `--tp_degree` argument no longer exists, so both baseline plans advertised in run_gemini.sh crash immediately with `AttributeError: 'Namespace' object has no attribute 'tp_degree'`. The demo has no tensor parallelism, so drop the stale assert. test_ci.sh only exercises the CAI_* plans, which is why this went unnoticed. --- examples/language/gpt/gemini/train_gpt_demo.py | 1 - 1 file changed, 1 deletion(-) diff --git a/examples/language/gpt/gemini/train_gpt_demo.py b/examples/language/gpt/gemini/train_gpt_demo.py index cb5d2c32ca6f..3501942f174b 100644 --- a/examples/language/gpt/gemini/train_gpt_demo.py +++ b/examples/language/gpt/gemini/train_gpt_demo.py @@ -176,7 +176,6 @@ def main(): logger.info(get_mem_info(prefix="After init optim, "), ranks=[0]) elif args.distplan.startswith("Pytorch"): - assert args.tp_degree == 1, "The degree of TP should be 1 for DDP examples." model = model_builder(args.model_type)(checkpoint=True).cuda() plugin = TorchDDPPlugin() if args.distplan.endswith("DDP"): From e7a3aa0e161a6538a5f490a62a52c36a3b1fabaf Mon Sep 17 00:00:00 2001 From: "pre-commit-ci[bot]" <66853113+pre-commit-ci[bot]@users.noreply.github.com> Date: Wed, 14 Oct 2026 01:05:08 +0000 Subject: [PATCH 2/2] [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --- tests/test_checkpoint_io/test_gemini_checkpoint_io.py | 2 +- tests/test_zero/test_gemini/test_optim.py | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/tests/test_checkpoint_io/test_gemini_checkpoint_io.py b/tests/test_checkpoint_io/test_gemini_checkpoint_io.py index ba60de0bd1c1..3649b18ec8bb 100644 --- a/tests/test_checkpoint_io/test_gemini_checkpoint_io.py +++ b/tests/test_checkpoint_io/test_gemini_checkpoint_io.py @@ -23,7 +23,7 @@ ASYNC_MODES = [False, True] if HAS_TENSORNVME else [False] try: - import apex.amp + pass APEX_AVAILABLE = True except ImportError: diff --git a/tests/test_zero/test_gemini/test_optim.py b/tests/test_zero/test_gemini/test_optim.py index b89c591882d6..a0970e43db9e 100644 --- a/tests/test_zero/test_gemini/test_optim.py +++ b/tests/test_zero/test_gemini/test_optim.py @@ -15,7 +15,7 @@ from tests.kit.model_zoo import model_zoo, run_fwd_bwd try: - import apex.amp + pass APEX_AMP_AVAILABLE = True except ImportError: