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

9
tests/basic_correctness/test_basic_correctness.py
View File

@@ -12,7 +12,6 @@ import pytest
import torch
from vllm import LLM, envs
from vllm.platforms import current_platform
from vllm.v1.engine.llm_engine import LLMEngine as LLMEngineV1
from ..conftest import HfRunner, VllmRunner
@@ -78,11 +77,7 @@ def test_models(
"VLLM_USE_V1") and envs.VLLM_USE_V1:
pytest.skip("enable_prompt_embeds is not supported in v1.")
if backend == "FLASHINFER" and current_platform.is_rocm():
pytest.skip("Flashinfer does not support ROCm/HIP.")
if backend in ("XFORMERS",
"FLASHINFER") and model == "google/gemma-2-2b-it":
if backend == "XFORMERS" and model == "google/gemma-2-2b-it":
pytest.skip(
f"{backend} does not support gemma2 with full context length.")
@@ -141,8 +136,6 @@ def test_models(
("meta-llama/Llama-3.2-1B-Instruct", "mp", "", "L4", {}),
("distilbert/distilgpt2", "ray", "", "A100", {}),
("distilbert/distilgpt2", "mp", "", "A100", {}),
("distilbert/distilgpt2", "mp", "FLASHINFER", "A100", {}),
("meta-llama/Meta-Llama-3-8B", "ray", "FLASHINFER", "A100", {}),
])
@pytest.mark.parametrize("enable_prompt_embeds", [True, False])
def test_models_distributed(