[CI][ROCm] Skip unsupported test cases on ROCm (#46160)

Signed-off-by: Felix Marty <Felix.Marty@amd.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

tests/quantization/test_compressed_tensors.py

@@ -5,6 +5,7 @@
 Run `pytest tests/quantization/test_compressed_tensors.py`.
 """
 
+from contextlib import contextmanager
 from unittest.mock import Mock
 
 import pytest
@@ -370,6 +371,27 @@ def test_compressed_tensors_kv_cache_fp8_per_attn_head(vllm_runner):
         assert output
 
 
+@contextmanager
+def _nvfp4_marlin_error_context(model, capfd):
+    is_rocm_and_unsupported = (
+        model == "nm-testing/TinyLlama-1.1B-Chat-v1.0-NVFP4A16"
+        and current_platform.is_rocm()
+    )
+
+    if is_rocm_and_unsupported:
+        expected_error = (
+            "ValueError: Forced NVFP4 kernel MarlinNvFp4LinearKernel is not "
+            "supported: Marlin FP4 not available"
+        )
+        with pytest.raises(RuntimeError, match="Engine core initialization failed"):
+            yield
+
+        captured = capfd.readouterr()
+        assert expected_error in captured.out + captured.err
+    else:
+        yield
+
+
 @pytest.mark.parametrize(
     "args",
     [
@@ -377,9 +399,12 @@ def test_compressed_tensors_kv_cache_fp8_per_attn_head(vllm_runner):
         ("nm-testing/TinyLlama-1.1B-Chat-v1.0-NVFP4", False),
     ],
 )
-def test_compressed_tensors_nvfp4(vllm_runner, args):
+def test_compressed_tensors_nvfp4(vllm_runner, args, capfd):
     model, use_a16 = args
-    with vllm_runner(model, enforce_eager=True) as llm:
+    with (
+        _nvfp4_marlin_error_context(model, capfd),
+        vllm_runner(model, enforce_eager=True) as llm,
+    ):
 
         def check_model(model):
             layer = model.model.layers[0]

tests/quantization/test_modelopt.py

@@ -25,6 +25,7 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,
 )
+from vllm.platforms import current_platform
 
 
 @pytest.fixture(scope="function", autouse=True)
@@ -467,6 +468,12 @@ def test_modelopt_mixed_precision_dispatches_w4a16_layer(
     from vllm.model_executor.layers.linear import LinearBase
     from vllm.model_executor.layers.quantization import modelopt as m
 
+    if (
+        expected_linear_cls_name == "ModelOptNvFp4W4A16LinearMethod"
+        and current_platform.is_rocm()
+    ):
+        pytest.skip("ModelOptNvFp4W4A16LinearMethod is not supported with rocm")
+
     hf_quant_config: dict[str, Any] = {
         "quantization": {
             "quant_algo": "MIXED_PRECISION",