Skip to content

[https://nvbugs/6566765][fix] Release Qwen MoE CUDA memory between tests #17641

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

Open
BowenFu wants to merge 1 commit into
base: main
Choose a base branch
from
+22 −3
Open

[https://nvbugs/6566765][fix] Release Qwen MoE CUDA memory between tests #17641

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
3 changes: 0 additions & 3 deletions tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -355,9 +355,6 @@ unittest/_torch/attention/test_attention_backends.py::test_attention_backend[dee
unittest/_torch/executor/test_overlap_scheduler.py::test_overlap_scheduler_consistency[no_reuse-cpp_scheduler-TorchSampler] SKIP (https://nvbugs/6561559)
unittest/_torch/misc/test_autotuner.py::test_cutedsl_nvfp4_heuristic_matches_full_sweep SKIP (https://nvbugs/6490028)
unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_31b_dummy SKIP (https://nvbugs/6607482)
unittest/_torch/modeling/test_modeling_qwen_moe.py::TestQwenMoe::test_qwen_moe_allclose_to_hf[backend:trtllm-use_cuda_graph:False] SKIP (https://nvbugs/6566765)
unittest/_torch/modeling/test_modeling_qwen_moe.py::TestQwenMoe::test_qwen_moe_allclose_to_hf[backend:trtllm-use_cuda_graph:True] SKIP (https://nvbugs/6575012)
unittest/_torch/modeling/test_modeling_qwen_moe.py::TestQwenMoe::test_qwen_moe_allclose_to_hf[backend:vanilla-use_cuda_graph:False] SKIP (https://nvbugs/6566765)
unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k "TRTLLM" SKIP (https://nvbugs/6602176)
unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend[act=Relu2-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize] SKIP (https://nvbugs/5989912)
unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k "MEGAMOE_CUTEDSL" SKIP (https://nvbugs/6601578)
Expand Down
22 changes: 22 additions & 0 deletions tests/unittest/_torch/modeling/test_modeling_qwen_moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,7 @@
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

import gc
import unittest
from copy import deepcopy
from dataclasses import dataclass
Expand Down Expand Up @@ -67,6 +71,24 @@ def __repr__(self) -> str:

class TestQwenMoe(unittest.TestCase):

@staticmethod
def _release_cuda_memory():
gc.collect()
torch.cuda.empty_cache()

@classmethod
def setUpClass(cls):
super().setUpClass()
# The A30 suite runs every Qwen modeling test in one process. Clear
# cyclic model references left by earlier classes before allocating
# both the HF and TensorRT-LLM models used by these comparisons.
cls._release_cuda_memory()

def tearDown(self):
# Do not leave either model alive for the next parameterized case.
self._release_cuda_memory()
super().tearDown()

@parameterized.expand([None, "FP8", "NVFP4"])
def test_qwen_moe_sanity(self, quant_algo):
config_dict = deepcopy(QWEN2_57B_A14B_CONFIG)
Expand Down
Loading