refactor: remove implicit redundant checks that obfuscate installation errors in test collection

Author: begumcig

tests/algorithms/test_combinations.py

@@ -60,7 +60,7 @@ def prepare_smash_config(self, smash_config: SmashConfig, device: str) -> None:
         ("flux_tiny_random", ["fora", "torch_compile"], False, 'cmmd'),
         ("flux_tiny_random", ["fora", "stable_fast"], False, 'cmmd'),
         ("tiny_janus", ["hqq", "torch_compile"], False, 'cmmd'),
-        pytest.param("flux_tiny", ["fora", "flash_attn3", "torch_compile"], False, 'cmmd', marks=pytest.mark.high_vram),
+        pytest.param("flux_tiny", ["fora", "flash_attn3", "torch_compile"], False, 'cmmd', marks=pytest.mark.high_gpu),
     ],
     indirect=["model_fixture"],
     ids=lambda val: "+".join(val) if isinstance(val, list) else None,

tests/algorithms/testers/cgenerate.py

@@ -1,9 +1,12 @@
+import pytest
+
 from pruna import PrunaModel
 from pruna.algorithms.c_translate import CGenerate
 
 from .base_tester import AlgorithmTesterBase
 
 
+@pytest.mark.requires_whisper
 class TestCGenerate(AlgorithmTesterBase):
     """Test the c_generate algorithm."""
 

tests/algorithms/testers/flash_attn3.py

@@ -5,7 +5,7 @@
 from .base_tester import AlgorithmTesterBase
 
 
-@pytest.mark.high_vram
+@pytest.mark.high_gpu
 class TestFlashAttn3(AlgorithmTesterBase):
     """Test the flash attention 3 kernel."""
 

tests/algorithms/testers/gptq.py

@@ -7,7 +7,7 @@
 
 
 @pytest.mark.slow
-@pytest.mark.high_vram
+@pytest.mark.high_gpu
 @pytest.mark.requires_gptq
 class TestGPTQ(AlgorithmTesterBase):
     """Test the GPTQ quantizer."""

tests/algorithms/testers/ipex_llm.py

@@ -5,7 +5,6 @@
 from .base_tester import AlgorithmTesterBase
 
 
-@pytest.mark.high_cpu
 @pytest.mark.requires_intel
 class TestIPEXLLM(AlgorithmTesterBase):
     """Test the IPEX LLM algorithm."""

tests/algorithms/testers/moe_kernel_tuner.py

@@ -36,7 +36,6 @@ def post_smash_hook(self, model: PrunaModel) -> None:
 
     def _resolve_hf_cache_config_path(self) -> Path:
         """Read the saved artifact and compute the expected HF cache config path."""
-
         imported_packages = MoeKernelTuner().import_algorithm_packages()
 
         smash_cfg = SmashConfig()

tests/algorithms/testers/sage_attn.py

@@ -5,7 +5,7 @@
 from .base_tester import AlgorithmTesterBase
 
 
-@pytest.mark.high_vram
+@pytest.mark.high_gpu
 class TestSageAttn(AlgorithmTesterBase):
     """Test the sage attention kernel."""
 

tests/algorithms/testers/tti_inplace_perp.py

@@ -27,7 +27,7 @@ def assert_no_nan_values(module: Any) -> None:
 
 # Our nightlies machine does not support efficient attention mechanisms and causes OOM errors with this test.
 # This test do pass on modern architectures.
-@pytest.mark.high_vram
+@pytest.mark.high_gpu
 @pytest.mark.slow
 class TestTTIInPlacePerp(AlgorithmTesterBase):
     """Test the TTI InPlace Perp recovery algorithm."""

tests/algorithms/testers/tti_lora.py

@@ -27,7 +27,7 @@ def assert_no_nan_values(module: Any) -> None:
 
 # Our nightlies machine does not support efficient attention mechanisms and causes OOM errors with this test.
 # This test do pass on modern architectures.
-@pytest.mark.high_vram
+@pytest.mark.high_gpu
 @pytest.mark.slow
 class TestTTILoRA(AlgorithmTesterBase):
     """Test the TTI LoRA recovery algorithm."""

tests/algorithms/testers/tti_perp.py

@@ -26,7 +26,7 @@ def assert_no_nan_values(module: Any) -> None:
 
 
 @pytest.mark.slow
-@pytest.mark.high_vram
+@pytest.mark.high_gpu
 class TestTTIPerp(AlgorithmTesterBase):
     """Test the TTI Perp recovery algorithm."""