[Sampler] Adapt to FlashInfer 0.2.3 sampler API (#15777)

Signed-off-by: Bowen Wang <abmfy@icloud.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Bowen Wang
2025-05-16 15:14:03 -07:00
committed by GitHub
parent aef94c6d07
commit 7fdfa01530
7 changed files with 122 additions and 88 deletions

View File

@@ -647,6 +647,8 @@ def test_flashinfer_fallback(seed: int, device: str):
if not envs.VLLM_USE_FLASHINFER_SAMPLER:
pytest.skip("Flashinfer sampler is disabled")
pytest.skip("After FlashInfer 0.2.3, sampling will never fail")
set_random_seed(seed)
torch.set_default_device(device)
batch_size = random.randint(1, 256)