Merge branch 'main' into dev/vllm-align

tests/diffusion/layers/test_activation.py

@@ -5,7 +5,7 @@ import pytest
 import torch
 import torch.nn.functional as F
 
-from tests.helpers.mark import hardware_test
+from tests.helpers.mark import hardware_marks
 from vllm_omni.diffusion.layers.activation import SiluAndMul
 
 pytestmark = [pytest.mark.core_model, pytest.mark.diffusion]
@@ -19,9 +19,17 @@ def test_silu_and_mul_native_matches_packed_reference() -> None:
     torch.testing.assert_close(SiluAndMul().forward_native(packed), F.silu(gate) * up)
 
 
-@hardware_test(res={"npu": "A3"}, num_cards=1)
-def test_silu_and_mul_npu_matches_packed_reference() -> None:
-    packed = torch.randn(257, 256, device="npu", dtype=torch.bfloat16)
+@pytest.mark.parametrize(
+    "dtype",
+    [
+        pytest.param(torch.bfloat16, marks=hardware_marks(res={"npu": "A2"}, num_cards=1), id="A2-bf16"),
+        pytest.param(torch.bfloat16, marks=hardware_marks(res={"npu": "A3"}, num_cards=1), id="A3-bf16"),
+        pytest.param(torch.bfloat16, marks=hardware_marks(res={"npu": "A5"}, num_cards=1), id="A5-bf16"),
+        pytest.param(torch.float16, marks=hardware_marks(res={"npu": "310P"}, num_cards=1), id="310P-fp16"),
+    ],
+)
+def test_silu_and_mul_npu_matches_packed_reference(dtype: torch.dtype) -> None:
+    packed = torch.randn(257, 256, device="npu", dtype=dtype)
     gate, up = packed.chunk(2, dim=-1)
 
     torch.testing.assert_close(

tests/diffusion/models/ltx2/test_ltx2_ops_platform.py

@@ -37,8 +37,10 @@ def clear_platform_cache():
         (90, True, True, True, True),
         (90, True, False, True, False),
         (90, False, True, False, False),
-        (100, True, True, False, False),
-        (103, True, True, False, False),
+        (100, True, True, True, False),
+        (100, True, False, True, False),
+        (103, True, True, True, False),
+        (103, True, False, True, False),
         (110, True, True, False, False),
         (120, True, True, False, False),
         (121, True, True, False, False),

tests/diffusion/models/ltx2/test_ltx2_vocoder_cuda.py

@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 
-"""CUDA regression tests for LTX-2 BWE vocoder precision."""
+"""GPU regression tests for LTX-2 BWE vocoder precision."""
 
 import pytest
 import torch
@@ -14,10 +14,13 @@ pytestmark = [pytest.mark.core_model, pytest.mark.diffusion]
 class _BWEConvVocoder(torch.nn.Module):
     def __init__(self):
         super().__init__()
+        # Match the production LTX vocoder's conv_pre geometry. The previous
+        # single-channel 1x1 probe intermittently had no MIOpen algorithm.
         self.conv = torch.nn.Conv1d(
-            1,
-            1,
-            kernel_size=1,
+            128,
+            1024,
+            kernel_size=7,
+            padding=3,
             bias=False,
             device="cuda",
             dtype=torch.bfloat16,
@@ -33,12 +36,12 @@ class _BWEConvVocoder(torch.nn.Module):
         return output
 
 
-@hardware_test(res={"cuda": "L4"}, num_cards=1)
+@hardware_test(res={"cuda": "L4", "rocm": "MI325"}, num_cards=1)
 def test_ltx_bwe_vocoder_runs_real_cuda_autocast_in_fp32():
     from vllm_omni.diffusion.models.ltx2.ltx2_runtime import _run_ltx_vocoder
 
     vocoder = _BWEConvVocoder()
-    generated_mel = torch.ones((1, 1, 4), device="cuda", dtype=torch.bfloat16)
+    generated_mel = torch.ones((1, 128, 64), device="cuda", dtype=torch.bfloat16)
 
     output = _run_ltx_vocoder(vocoder, generated_mel)
 

tests/diffusion/quantization/test_quantization_fp8.py

@@ -300,7 +300,8 @@ def test_single_stage_zimage_fp8_uses_less_memory():
     assert mem_fp8 < mem_bf16, f"FP8 ({mem_fp8:.2f} GiB) should use less memory than BF16 ({mem_bf16:.2f} GiB)"
 
 
-@hardware_test(res={"cuda": "L4"})
+@hardware_test(res={"cuda": "L4", "xpu": "B60"})
+@pytest.mark.advanced_model
 def test_single_stage_qwen_image_fp8():
     """Qwen-Image (random weights) with FP8 generates valid images."""
     model = "riverclouds/qwen_image_random"