EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

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

Browse Source 
### 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
Download Patch File Download Diff File Expand all files Collapse all files

12
tests/e2e/singlecard/test_aclgraph_accuracy.py
View File

@@ -64,6 +64,7 @@ def test_models_output_between_eager_and_aclgraph(
with VllmRunner(
model,
max_model_len=1024,
cudagraph_capture_sizes=[1, 2, 4, 8],
quantization="ascend",
) as runner:
vllm_aclgraph_outputs = runner.model.generate(
@@ -72,6 +73,7 @@ def test_models_output_between_eager_and_aclgraph(
with VllmRunner(
model,
max_model_len=1024,
cudagraph_capture_sizes=[1, 2, 4, 8],
) as runner:
vllm_aclgraph_outputs = runner.model.generate(
prompts, sampling_params)
@@ -151,7 +153,10 @@ def test_models_output_between_eager_and_full_decode_only(
with VllmRunner(
model,
max_model_len=1024,
compilation_config={"cudagraph_mode": "FULL_DECODE_ONLY"},
compilation_config={
"cudagraph_capture_sizes": [4, 8, 32, 64],
"cudagraph_mode": "FULL_DECODE_ONLY"
},
quantization="ascend",
) as runner:
vllm_aclgraph_outputs = runner.model.generate(
@@ -245,7 +250,10 @@ def test_models_output_between_eager_and_fullgraph_npugraph_ex(
with VllmRunner(
model,
max_model_len=1024,
compilation_config={"cudagraph_mode": "FULL_DECODE_ONLY"},
compilation_config={
"cudagraph_capture_sizes": [4, 8, 32, 64],
"cudagraph_mode": "FULL_DECODE_ONLY"
},
additional_config={"enable_npugraph_ex": True},
quantization="ascend",
) as runner: