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

[Attention] Update tests to remove deprecated env vars (#30563)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Matthew Bonanni
2025-12-17 12:49:59 -05:00
committed by GitHub
parent 9ca8cb38fd
commit 7eb6cb6c18
34 changed files with 580 additions and 447 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
tests/v1/e2e/test_cascade_attention.py
View File

@@ -10,7 +10,7 @@ from ...utils import create_new_process_for_each_test
@create_new_process_for_each_test()
@pytest.mark.parametrize("attn_backend", ["FLASH_ATTN", "FLASHINFER"])
def test_cascade_attention(example_system_message, monkeypatch, attn_backend):
def test_cascade_attention(example_system_message, attn_backend):
prompt = "\n<User>: Implement fibonacci sequence in Python.\n<Claude>:"
if attn_backend == "FLASHINFER":
@@ -19,19 +19,18 @@ def test_cascade_attention(example_system_message, monkeypatch, attn_backend):
"needs investigation. See issue #25679."
)
with monkeypatch.context() as m:
m.setenv("VLLM_ATTENTION_BACKEND", attn_backend)
llm = LLM(
model="Qwen/Qwen2-1.5B-Instruct", attention_config={"backend": attn_backend}
)
sampling_params = SamplingParams(temperature=0.0, max_tokens=100)
llm = LLM(model="Qwen/Qwen2-1.5B-Instruct")
sampling_params = SamplingParams(temperature=0.0, max_tokens=100)
# No cascade attention.
single_prompt = [example_system_message + prompt]
responses = llm.generate(single_prompt, sampling_params)
ref_output = responses[0].outputs[0].text
# No cascade attention.
single_prompt = [example_system_message + prompt]
responses = llm.generate(single_prompt, sampling_params)
ref_output = responses[0].outputs[0].text
# (Probably) Use cascade attention.
prompts = [example_system_message + prompt] * 64
responses = llm.generate(prompts, sampling_params)
for response in responses:
assert response.outputs[0].text == ref_output
# (Probably) Use cascade attention.
prompts = [example_system_message + prompt] * 64
responses = llm.generate(prompts, sampling_params)
for response in responses:
assert response.outputs[0].text == ref_output