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

ACLgraph enable: Test cases revisions for all features (#3388)

Browse Source 
### What this PR does / why we need it?
This PR revise the test cases of various features on the warehouse which
add the enablement of aclgraph to the test cases.

### Does this PR introduce _any_ user-facing change?
no

### How was this patch tested?
ut

- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

Signed-off-by: lilinsiman <lilinsiman@gmail.com>
This commit is contained in:
lilinsiman
2025-10-17 17:15:19 +08:00
committed by GitHub
parent bf87606932
commit 1b424fb7f1
17 changed files with 34 additions and 117 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py
View File

@@ -71,7 +71,7 @@ def test_ngram_correctness(
should be the same when using ngram speculative decoding.
'''
pytest.skip("Not current support for the test.")
ref_llm = LLM(model=model_name, max_model_len=1024, enforce_eager=True)
ref_llm = LLM(model=model_name, max_model_len=1024, enforce_eager=False)
ref_outputs = ref_llm.chat(test_prompts, sampling_config)
del ref_llm
with VllmRunner(model_name,
@@ -82,7 +82,7 @@ def test_ngram_correctness(
"num_speculative_tokens": 3,
},
max_model_len=1024,
enforce_eager=True) as runner:
enforce_eager=False) as runner:
spec_outputs = runner.model.chat(test_prompts, sampling_config)
matches = 0
misses = 0
@@ -111,7 +111,7 @@ def test_eagle_correctness(
should be the same when using eagle speculative decoding.
'''
ref_llm = LLM(model=model_name, max_model_len=2048, enforce_eager=True)
ref_llm = LLM(model=model_name, max_model_len=2048, enforce_eager=False)
ref_outputs = ref_llm.chat(test_prompts, sampling_config)
del ref_llm
@@ -129,7 +129,7 @@ def test_eagle_correctness(
"max_model_len": 128,
},
max_model_len=128,
enforce_eager=True,
enforce_eager=False,
) as runner:
spec_outputs = runner.model.chat(test_prompts, sampling_config)

10
tests/e2e/singlecard/test_ascend_scheduler.py
View File

@@ -9,7 +9,8 @@ from tests.e2e.model_utils import check_outputs_equal
MODEL = "Qwen/Qwen3-0.6B"
def test_concurrent_partial_prefill():
@pytest.mark.parametrize("enforce_eager", [True, False])
def test_concurrent_partial_prefill(enforce_eager):
with VllmRunner(MODEL,
additional_config={
'ascend_scheduler_config': {
@@ -18,7 +19,7 @@ def test_concurrent_partial_prefill():
},
max_num_seqs=3,
max_num_batched_tokens=2048,
enforce_eager=True,
enforce_eager=enforce_eager,
max_model_len=2048,
gpu_memory_utilization=0.7) as vllm_model:
outputs = vllm_model.model.generate(["Hello my name is Robert and I"] *
@@ -28,7 +29,8 @@ def test_concurrent_partial_prefill():
assert len(output.outputs) == 1
def test_prefix_cache_stats_is_recorded():
@pytest.mark.parametrize("enforce_eager", [True, False])
def test_prefix_cache_stats_is_recorded(enforce_eager):
with VllmRunner(MODEL,
additional_config={
'ascend_scheduler_config': {
@@ -37,7 +39,7 @@ def test_prefix_cache_stats_is_recorded():
},
max_num_seqs=3,
max_num_batched_tokens=2048,
enforce_eager=True,
enforce_eager=enforce_eager,
max_model_len=2048,
gpu_memory_utilization=0.7) as vllm_model:
# 17 tokens will make sure first 16 tokens are cached in a block

2
tests/e2e/singlecard/test_camem.py
View File

@@ -74,7 +74,7 @@ def test_end_to_end():
sampling_params = SamplingParams(temperature=0, max_tokens=10)
with VllmRunner("Qwen/Qwen3-0.6B",
enforce_eager=True,
enforce_eager=False,
enable_sleep_mode=True) as runner:
output = runner.model.generate(prompt, sampling_params)

7
tests/e2e/singlecard/test_chunked.py
View File

@@ -43,12 +43,13 @@ def test_models(
temperature=0.0,
)
with VllmRunner(model, long_prefill_token_threshold=20,
enforce_eager=True) as vllm_model:
with VllmRunner(model,
long_prefill_token_threshold=20,
enforce_eager=False) as vllm_model:
output1 = vllm_model.generate(prompts, sampling_params)
with VllmRunner(model,
enforce_eager=True,
enforce_eager=False,
additional_config={
'ascend_scheduler_config': {
'enabled': True

2
tests/e2e/singlecard/test_embedding.py
View File

@@ -29,7 +29,7 @@ def test_embed_models_correctness():
with VllmRunner(
model_name,
task="embed",
enforce_eager=True,
enforce_eager=False,
) as vllm_runner:
vllm_outputs = vllm_runner.encode(queries)

2
tests/e2e/singlecard/test_ilama_lora.py
View File

@@ -51,7 +51,7 @@ def test_ilama_lora(ilama_lora_files):
max_loras=4,
max_model_len=1024,
max_num_seqs=16,
enforce_eager=True) as vllm_model:
enforce_eager=False) as vllm_model:
output1 = do_sample(vllm_model.model, ilama_lora_files, lora_id=1)
for i in range(len(EXPECTED_LORA_OUTPUT)):

2
tests/e2e/singlecard/test_quantization.py
View File

@@ -28,7 +28,7 @@ def test_quant_W8A8():
with VllmRunner(
snapshot_download("vllm-ascend/Qwen2.5-0.5B-Instruct-W8A8"),
max_model_len=8192,
enforce_eager=True,
enforce_eager=False,
gpu_memory_utilization=0.7,
quantization="ascend",
) as vllm_model:

2
tests/e2e/singlecard/test_vlm.py
View File

@@ -46,7 +46,7 @@ def test_multimodal_vl(prompt_template):
"max_pixels": 1280 * 28 * 28,
"fps": 1,
},
enforce_eager=True) as vllm_model:
enforce_eager=False) as vllm_model:
outputs = vllm_model.generate_greedy(prompts=prompts,
images=images,
max_tokens=64)