biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[V0 Deprecation] Remove V0 FlashInfer attention backend (#22776)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon
2025-08-18 19:54:16 -07:00
committed by GitHub
parent 6603288736
commit 14006840ea
8 changed files with 9 additions and 1133 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/models/quantization/test_fp8.py
View File

@@ -32,7 +32,7 @@ from ..utils import check_logprobs_close
# Due to low-precision numerical divergence, we only test logprob of 4 tokens
@pytest.mark.parametrize("max_tokens", [4])
@pytest.mark.parametrize("enforce_eager", [True])
@pytest.mark.parametrize("backend", ["FLASH_ATTN", "XFORMERS", "FLASHINFER"])
@pytest.mark.parametrize("backend", ["FLASH_ATTN", "XFORMERS"])
# NOTE: Increasing this in this suite will fail CI because we currently cannot
# reset distributed env properly. Use a value > 1 just when you test.
@pytest.mark.parametrize("tensor_parallel_size", [1])
@@ -57,9 +57,6 @@ def test_models(
numerical sensitive kernels.
"""
if backend == "FLASHINFER" and current_platform.is_rocm():
pytest.skip("Flashinfer does not support ROCm/HIP.")
if kv_cache_dtype == "fp8_e5m2" and current_platform.is_rocm():
pytest.skip(
f"{kv_cache_dtype} is currently not supported on ROCm/HIP.")