Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[PyTorch] Fix non-reentrant checkpointing test #685

Closed
wants to merge 1 commit into from
Closed

[PyTorch] Fix non-reentrant checkpointing test #685

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion qa/L0_pytorch_unittest/test.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@ set -e
pip install pytest==6.2.5 onnxruntime==1.13.1
pytest -v -s $TE_PATH/tests/pytorch/test_sanity.py
pytest -v -s $TE_PATH/tests/pytorch/test_deferred_init.py
PYTORCH_JIT=0 NVTE_ALLOW_NONDETERMINISTIC_ALGO=0 pytest -v -s $TE_PATH/tests/pytorch/test_numerics.py
NVTE_TORCH_COMPILE=0 PYTORCH_JIT=0 NVTE_ALLOW_NONDETERMINISTIC_ALGO=0 pytest -v -s $TE_PATH/tests/pytorch/test_numerics.py
pytest -v -s $TE_PATH/tests/pytorch/test_jit.py
pytest -v -s $TE_PATH/tests/pytorch/fused_attn/test_fused_attn.py
pytest -v -s $TE_PATH/tests/pytorch/test_fused_rope.py
Expand Down
8 changes: 0 additions & 8 deletions tests/pytorch/test_numerics.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -558,19 +558,11 @@ def test_gpt_full_activation_recompute(dtype, bs, model, fp8, fp8_model_params,

config = model_configs[model]

if not use_reentrant:
# Non-reentrant checkpoint becomes non-deterministic with bias+GELU fusion
os.environ["NVTE_BIAS_GELU_NVFUSION"] = "0"

outputs, names = _test_e2e_full_recompute(bs, dtype, config, fp8, fp8_model_params,
recompute=False, use_reentrant=use_reentrant)
outputs_recompute, _ = _test_e2e_full_recompute(bs, dtype, config, fp8, fp8_model_params,
recompute=True, use_reentrant=use_reentrant)

if not use_reentrant:
# Reset bias+GELU fusion flag to avoid contaminating other tests
del os.environ["NVTE_BIAS_GELU_NVFUSION"]

assert_all_equal(outputs, outputs_recompute, names=names)


Expand Down
5 changes: 0 additions & 5 deletions transformer_engine/pytorch/module/layernorm_mlp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1416,11 +1416,6 @@ def forward(
is_first_microbatch
)

# Disable bias_gelu_nvfusion for determinism checkpointing in non-reentrant mode
if (self.bias_gelu_nvfusion
and not use_reentrant_activation_recompute()):
self.bias_gelu_nvfusion = False

from ..cpu_offload import CPUOffloadEnabled

if torch.is_grad_enabled():
Expand Down
Loading