[Misc] Fix Current vLLM config is not set. warnings, assert to avoid issues in the future (#31747)

Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Signed-off-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>

tests/kernels/core/test_activation.py

@@ -45,6 +45,7 @@ CUDA_DEVICES = [f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 e
 @pytest.mark.parametrize("device", CUDA_DEVICES)
 @torch.inference_mode()
 def test_act_and_mul(
+    default_vllm_config,
     activation: str,
     num_tokens: int,
     d: int,
@@ -122,6 +123,7 @@ def test_act_and_mul(
 @pytest.mark.parametrize("device", CUDA_DEVICES)
 @torch.inference_mode()
 def test_activation(
+    default_vllm_config,
     activation: type[torch.nn.Module],
     num_tokens: int,
     d: int,

tests/kernels/core/test_fused_qk_norm_rope.py

@@ -57,6 +57,7 @@ def _apply_qk_norm_rope(
 @pytest.mark.parametrize("rotary_ratio", [1.0, 0.5, 0.25])
 @torch.inference_mode()
 def test_fused_qk_norm_rope_matches_reference(
+    default_vllm_config,
     device: str,
     dtype: torch.dtype,
     is_neox: bool,

tests/kernels/core/test_fused_quant_layernorm.py

@@ -147,6 +147,7 @@ def ops_impl(
 @pytest.mark.parametrize("device", CUDA_DEVICES)
 @torch.inference_mode()
 def test_rms_norm(
+    default_vllm_config,
     num_tokens: int,
     hidden_size: int,
     add_residual: bool,

tests/kernels/core/test_layernorm.py

@@ -26,6 +26,7 @@ CUDA_DEVICES = [f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 e
 @pytest.mark.parametrize("strided_input", [False, True])
 @torch.inference_mode()
 def test_rms_norm(
+    default_vllm_config,
     num_tokens: int,
     hidden_size: int,
     add_residual: bool,

tests/kernels/core/test_mrope.py

@@ -90,6 +90,7 @@ num_tokens_list = [11, 8192]
 @pytest.mark.parametrize("dtype", [torch.bfloat16])
 @pytest.mark.parametrize("num_tokens", num_tokens_list)
 def test_mrope(
+    default_vllm_config,
     model_name: str,
     model_info: MRoPETestInfo,
     tp_size: int,
@@ -159,6 +160,7 @@ def test_mrope(
 @pytest.mark.parametrize("dtype", [torch.bfloat16])
 @pytest.mark.parametrize("num_tokens", num_tokens_list)
 def test_mrope_torch_compile_tracing(
+    default_vllm_config,
     model_name: str,
     model_info: MRoPETestInfo,
     tp_size: int,

tests/kernels/core/test_pos_encoding.py

@@ -62,6 +62,7 @@ TENSORS_SHAPES_FN = [
 @pytest.mark.parametrize("use_key", USE_KEY)
 @torch.inference_mode()
 def test_rotary_embedding(
+    default_vllm_config,
     is_neox_style: bool,
     tensor_shape_fn: Callable[[int, int, int, int], tuple[int, ...]],
     batch_size: int,
@@ -123,7 +124,7 @@ def test_rotary_embedding(
 
 
 @torch.inference_mode()
-def test_rope_module_cache():
+def test_rope_module_cache(default_vllm_config):
     MAX_POSITIONS = [123, 1234]
     ROPE_THETAS = [10000, 1000000]
     ROPE_PARAMETERS = (

tests/kernels/core/test_rotary_embedding.py

@@ -36,6 +36,7 @@ def rotary_embedding_opcheck(
 @pytest.mark.parametrize("use_key", [True, False])
 @pytest.mark.parametrize("head_stride_is_contiguous", [True, False])
 def test_rotary_embedding_opcheck(
+    default_vllm_config,
     dist_init,
     device,
     max_position,