[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>
This commit is contained in:
Lucas Wilkinson
2026-01-08 15:20:49 -08:00
committed by GitHub
co-authored by Luka Govedič
parent 5d3b6097ad
commit 6cdf015c3c
48 changed files with 380 additions and 240 deletions
@@ -458,7 +458,7 @@ def test_flashinfer_trtllm_prefill_with_baseline(
)
def test_trtllm_attention_rejects_num_kv_heads_1() -> None:
def test_trtllm_attention_rejects_num_kv_heads_1(default_vllm_config) -> None:
"""Test that TRTLLM attention correctly rejects num_kv_heads=1.
When num_kv_heads=1 (MQA), the KV cache strides become degenerate
+3 -1
View File
@@ -36,7 +36,7 @@ if current_platform.is_rocm():
@pytest.mark.parametrize("device", devices)
def test_mha_attn_platform(device: str):
def test_mha_attn_platform(default_vllm_config, device: str):
"""
Test the attention selector between different platform and device.
"""
@@ -116,6 +116,7 @@ CUDA_DEVICES = ["cuda"]
@pytest.mark.parametrize("dtype", DTYPES)
@pytest.mark.parametrize("device", CUDA_DEVICES)
def test_mha_attn_forward(
default_vllm_config,
batch_size: int,
seq_len: int,
num_heads: int,
@@ -162,6 +163,7 @@ def test_mha_attn_forward(
@pytest.mark.parametrize("dtype", DTYPES)
@pytest.mark.parametrize("device", CUDA_DEVICES)
def test_mha_attn_varlen_forward(
default_vllm_config,
var_seq_len: list[int],
num_heads: int,
num_kv_heads: int,