[CI/Build] Share nightly CUDA YAML between H100/L4 and B200 (#7028)

Signed-off-by: wangyu <410167048@qq.com>
Signed-off-by: [Your Name] [Your Email]
Signed-off-by: [Your Name] <your.email@example.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: amy-why-3459 <wuhaiyan17@huawei.com>

tests/diffusion/distributed/test_cfg_parallel.py

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 """Tests for CFG (Classifier-Free Guidance) parallel functionality.
 
 CPU tests mock the CFG process group and validate mixin logic without GPUs.
@@ -32,8 +32,8 @@ from vllm_omni.diffusion.distributed.parallel_state import (
 )
 from vllm_omni.platforms import current_omni_platform
 
-_L4_TWO_GPU = hardware_marks(res={"cuda": "L4"}, num_cards=2)
-_L4_THREE_GPU = hardware_marks(res={"cuda": "L4"}, num_cards=3)
+_L4_TWO_GPU = hardware_marks(res={"cuda": ["L4", "B200"]}, num_cards=2)
+_L4_THREE_GPU = hardware_marks(res={"cuda": ["L4", "B200"]}, num_cards=3)
 
 
 def _set_random_seeds(seed: int) -> None:
@@ -1009,7 +1009,7 @@ def _run_multi_branch_gpu_parity(
 @pytest.mark.full_model
 @pytest.mark.diffusion
 @pytest.mark.parallel
-@hardware_test(res={"cuda": "L4"}, num_cards=2)
+@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=2)
 @pytest.mark.parametrize("batch_size", [2])
 @pytest.mark.parametrize("cfg_normalize", [False, True])
 @pytest.mark.parametrize("extra_kwargs", [None, {"step_i": 1}])

tests/diffusion/distributed/test_comm.py

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 """Tests for SeqAllToAll4D, SeqAllToAll5D, and RingComm communication primitives.
 
 CPU tests use gloo on CPU tensors (no GPU). Nightly parity tests run the same
@@ -24,8 +24,8 @@ from vllm_omni.diffusion.distributed.parallel_state import (
 )
 from vllm_omni.platforms import current_omni_platform
 
-_L4_TWO_GPU = hardware_marks(res={"cuda": "L4"}, num_cards=2)
-_L4_FOUR_GPU = hardware_marks(res={"cuda": "L4"}, num_cards=4)
+_L4_TWO_GPU = hardware_marks(res={"cuda": ["L4", "B200"]}, num_cards=2)
+_L4_FOUR_GPU = hardware_marks(res={"cuda": ["L4", "B200"]}, num_cards=4)
 
 DeviceKind = Literal["cpu", "cuda"]
 

tests/diffusion/distributed/test_expert_parallel.py

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 
 """
 System test for Expert Parallel (EP).

tests/diffusion/distributed/test_flux2_vae_patch_parallel.py

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 
 """FLUX.2-dev VAE patch-parallel correctness tests.
 

tests/diffusion/distributed/test_pipeline_parallel.py

@@ -31,9 +31,9 @@ from vllm_omni.diffusion.distributed.parallel_state import (
 from vllm_omni.diffusion.distributed.pipeline_parallel import AsyncLatents, PipelineParallelMixin
 from vllm_omni.platforms import current_omni_platform
 
-_L4_TWO_GPU = hardware_marks(res={"cuda": "L4"}, num_cards=2)
-_L4_THREE_GPU = hardware_marks(res={"cuda": "L4"}, num_cards=3)
-_L4_FOUR_GPU = hardware_marks(res={"cuda": "L4"}, num_cards=4)
+_L4_TWO_GPU = hardware_marks(res={"cuda": ["L4", "B200"]}, num_cards=2)
+_L4_THREE_GPU = hardware_marks(res={"cuda": ["L4", "B200"]}, num_cards=3)
+_L4_FOUR_GPU = hardware_marks(res={"cuda": ["L4", "B200"]}, num_cards=4)
 
 DeviceKind = Literal["cpu", "cuda"]
 _UNIT_MARKS = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]

tests/diffusion/distributed/test_sana_video_sp_numeric.py

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 """Real multi-process numeric equivalence gate for SANA-Video sequence parallel.
 
 Runs the native transformer directly (no pipeline, no topology validation) with
@@ -273,7 +273,7 @@ def _baseline_worker(
 @pytest.mark.full_model
 @pytest.mark.diffusion
 @pytest.mark.parallel
-@hardware_test(res={"cuda": "L4"}, num_cards=2)
+@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=2)
 @pytest.mark.parametrize("task", ["t2v", "i2v"])
 @pytest.mark.parametrize("num_frames", [21, 5])
 def test_sp2_matches_sp1_fp32(task: str, num_frames: int) -> None:
@@ -290,7 +290,7 @@ def test_sp2_matches_sp1_fp32(task: str, num_frames: int) -> None:
 @pytest.mark.full_model
 @pytest.mark.diffusion
 @pytest.mark.parallel
-@hardware_test(res={"cuda": "L4"}, num_cards=4)
+@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=4)
 @pytest.mark.parametrize("task", ["t2v", "i2v"])
 @pytest.mark.parametrize("num_frames", [21, 5])
 def test_sp4_matches_sp1_fp32(task: str, num_frames: int) -> None:
@@ -307,7 +307,7 @@ def test_sp4_matches_sp1_fp32(task: str, num_frames: int) -> None:
 @pytest.mark.full_model
 @pytest.mark.diffusion
 @pytest.mark.parallel
-@hardware_test(res={"cuda": "L4"}, num_cards=4)
+@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=4)
 def test_sp2_cfg2_matches_sp1_fp32() -> None:
     """Four processes: CFG branches with distinct inputs each run their own SP
     group; catches CFG/SP group crossover and deadlocks."""
@@ -324,7 +324,7 @@ def test_sp2_cfg2_matches_sp1_fp32() -> None:
 @pytest.mark.full_model
 @pytest.mark.diffusion
 @pytest.mark.parallel
-@hardware_test(res={"cuda": "L4"}, num_cards=4)
+@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=4)
 def test_tp2_sp2_matches_sp1_fp32() -> None:
     """Four processes: TP shards heads and channels while SP shards frames; the
     combined mesh must still reproduce the serial output."""
@@ -342,7 +342,7 @@ def test_tp2_sp2_matches_sp1_fp32() -> None:
 @pytest.mark.full_model
 @pytest.mark.diffusion
 @pytest.mark.parallel
-@hardware_test(res={"cuda": "L4"}, num_cards=2)
+@hardware_test(res={"cuda": ["L4", "B200"]}, num_cards=2)
 def test_sp2_bf16_drift_informational() -> None:
     """Records the bf16 reduction-reorder drift magnitude; the load-bearing
     numeric gate is the fp32 suite above."""

tests/diffusion/distributed/test_sequence_parallel.py

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 
 """
 System test for Sequence Parallel (SP) backends: Ulysses and Ring attention.

tests/diffusion/distributed/test_vae_decode_parallelism.py

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 
 """VAE patch-parallel decode parity tests for diffusion models.
 
@@ -145,7 +145,7 @@ def _run_generate(
 @pytest.mark.full_model
 @pytest.mark.diffusion
 @pytest.mark.parallel
-@hardware_test(res={"cuda": "H100"}, num_cards={"cuda": 4})
+@hardware_test(res={"cuda": ["H100", "B200"]}, num_cards={"cuda": 4})
 @pytest.mark.parametrize("model_case", MODEL_CASES)
 def test_vae_patch_parallel_tp2(model_case: dict[str, Any], tmp_path: Path):
     if current_omni_platform.is_npu():

tests/diffusion/distributed/test_wan_spatial_shard.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 import os
 from types import SimpleNamespace
 
@@ -538,7 +541,7 @@ def _spatial_shard_decode_worker(rank: int, split_dim: str, return_dict, master_
 @pytest.mark.full_model
 @pytest.mark.diffusion
 @pytest.mark.parallel
-@hardware_test(res={"cuda": "H100"}, num_cards=_SPATIAL_SHARD_WORLD_SIZE)
+@hardware_test(res={"cuda": ["H100", "B200"]}, num_cards=_SPATIAL_SHARD_WORLD_SIZE)
 @pytest.mark.parametrize("split_dim", ["height", "width"])
 def test_spatial_shard_decode_matches_reference(split_dim: str):
     manager = mp.get_context("spawn").Manager()