[E2E] Optimize the E2E test time. (#5294)

### What this PR does / why we need it?
Add cudagraph_capture_sizes for E2E CI test.

- vLLM version: release/v0.13.0
- vLLM main:
ad32e3e19c

Signed-off-by: menogrey <1299267905@qq.com>
This commit is contained in:
zhangyiming
2025-12-26 14:17:50 +08:00
committed by GitHub
parent 29d2fe653d
commit 45c5bcd962
22 changed files with 57 additions and 5 deletions

View File

@@ -37,7 +37,9 @@ def test_end_to_end():
prompt = "How are you?"
sampling_params = SamplingParams(temperature=0, max_tokens=10)
with VllmRunner("Qwen/Qwen3-0.6B", enable_sleep_mode=True) as runner:
with VllmRunner("Qwen/Qwen3-0.6B",
enable_sleep_mode=True,
cudagraph_capture_sizes=[1, 2, 4, 8]) as runner:
output = runner.model.generate(prompt, sampling_params)
# the benefit of `llm.sleep(level=2)` is mainly CPU memory usage,