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

[Lint]Style: Convert test/ to ruff format(Batch #5) (#6747)

Browse Source 
### What this PR does / why we need it?
| File Path |
| :--- |
| `tests/e2e/singlecard/compile/backend.py` |
| `tests/e2e/singlecard/compile/test_graphex_norm_quant_fusion.py` |
| `tests/e2e/singlecard/compile/test_graphex_qknorm_rope_fusion.py` |
| `tests/e2e/singlecard/compile/test_norm_quant_fusion.py` |
| `tests/e2e/singlecard/model_runner_v2/test_basic.py` |
| `tests/e2e/singlecard/test_aclgraph_accuracy.py` |
| `tests/e2e/singlecard/test_aclgraph_batch_invariant.py` |
| `tests/e2e/singlecard/test_aclgraph_mem.py` |
| `tests/e2e/singlecard/test_async_scheduling.py` |
| `tests/e2e/singlecard/test_auto_fit_max_mode_len.py` |
| `tests/e2e/singlecard/test_batch_invariant.py` |
| `tests/e2e/singlecard/test_camem.py` |
| `tests/e2e/singlecard/test_completion_with_prompt_embeds.py` |
| `tests/e2e/singlecard/test_cpu_offloading.py` |
| `tests/e2e/singlecard/test_guided_decoding.py` |
| `tests/e2e/singlecard/test_ilama_lora.py` |
| `tests/e2e/singlecard/test_llama32_lora.py` |
| `tests/e2e/singlecard/test_models.py` |
| `tests/e2e/singlecard/test_multistream_overlap_shared_expert.py` |
| `tests/e2e/singlecard/test_quantization.py` |
| `tests/e2e/singlecard/test_qwen3_multi_loras.py` |
| `tests/e2e/singlecard/test_sampler.py` |
| `tests/e2e/singlecard/test_vlm.py` |
| `tests/e2e/singlecard/test_xlite.py` |
| `tests/e2e/singlecard/utils.py` |

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.15.0
- vLLM main:
9562912cea

---------

Signed-off-by: MrZ20 <2609716663@qq.com>
This commit is contained in:
SILONG ZENG
2026-02-24 15:50:00 +08:00
committed by GitHub
parent 747484cb64
commit 62ea664aa7
26 changed files with 859 additions and 1052 deletions
Download Patch File Download Diff File Expand all files Collapse all files

104
tests/e2e/singlecard/test_async_scheduling.py
View File

@@ -15,8 +15,7 @@ from tests.e2e.model_utils import check_outputs_equal
MODEL = "Qwen/Qwen3-0.6B"
MTP_MODEL = "wemaster/deepseek_mtp_main_random_bf16"
first_prompt = ("The following numbers of the sequence " +
", ".join(str(i) for i in range(10)) + " are:")
first_prompt = "The following numbers of the sequence " + ", ".join(str(i) for i in range(10)) + " are:"
example_prompts = [
"Hello, my name is",
"The president of the United States is",
@@ -31,7 +30,9 @@ default_params = dict(
)
def test_without_spec_decoding(monkeypatch: pytest.MonkeyPatch, ):
def test_without_spec_decoding(
monkeypatch: pytest.MonkeyPatch,
):
"""Test consistency of combos of async scheduling, preemption,
uni/multiproc executor, prefill chunking."""
test_sampling_params: list[dict[str, Any]] = [
@@ -85,11 +86,11 @@ def run_tests(
# avoid precision errors
outputs: list[tuple[str, list, list]] = []
for n, (
test_preemption,
executor,
async_scheduling,
spec_config,
test_prefill_chunking,
test_preemption,
executor,
async_scheduling,
spec_config,
test_prefill_chunking,
) in enumerate(test_configs, 1):
test_str = f"{n}/{len(test_configs)}"
test_results = run_test(
@@ -105,21 +106,18 @@ def run_tests(
outputs.append(test_results)
baseline_config, baseline_tests, _ = outputs[0]
_, _, baseline_acceptances = next((o for o in outputs if o[2] is not None),
(None, None, None))
_, _, baseline_acceptances = next((o for o in outputs if o[2] is not None), (None, None, None))
print(
f"BASELINE: config=[{baseline_config}], accept_rates={baseline_acceptances}"
)
print(f"BASELINE: config=[{baseline_config}], accept_rates={baseline_acceptances}")
failure = None
for test_config, test_outputs, test_acceptance_rates in outputs[1:]:
for base_outs, base_acceptance_rate, test_outs, test_acceptance_rate, params in zip(
baseline_tests,
baseline_acceptances or repeat(None),
test_outputs,
test_acceptance_rates or repeat(None),
test_sampling_params,
baseline_tests,
baseline_acceptances or repeat(None),
test_outputs,
test_acceptance_rates or repeat(None),
test_sampling_params,
):
try:
check_outputs_equal(
@@ -129,21 +127,18 @@ def run_tests(
name_1=f"config=[{test_config}], params={params}",
)
if (base_acceptance_rate is not None
and test_acceptance_rate is not None):
if base_acceptance_rate is not None and test_acceptance_rate is not None:
if "spec_mml=None" in test_config:
assert (test_acceptance_rate > base_acceptance_rate
or test_acceptance_rate == pytest.approx(
base_acceptance_rate, rel=5e-2))
assert test_acceptance_rate > base_acceptance_rate or test_acceptance_rate == pytest.approx(
base_acceptance_rate, rel=5e-2
)
else:
# Currently the reported acceptance rate is expected to be
# lower when we sometimes skip drafting altogether.
assert test_acceptance_rate > 0.1
print(f"PASSED: config=[{test_config}], params={params}"
f" accept_rate={test_acceptance_rate}")
print(f"PASSED: config=[{test_config}], params={params} accept_rate={test_acceptance_rate}")
except AssertionError as e:
print(f"FAILED: config=[{test_config}], params={params}"
f" accept_rate={test_acceptance_rate}")
print(f"FAILED: config=[{test_config}], params={params} accept_rate={test_acceptance_rate}")
if failure is None:
failure = e
@@ -161,33 +156,35 @@ def run_test(
spec_config: dict[str, Any] | None,
test_prefill_chunking: bool,
):
os.environ['VLLM_WORKER_MULTIPROC_METHOD'] = 'spawn'
os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
spec_decoding = spec_config is not None
cache_arg: dict[str, Any] = (
# Force preemptions
dict(num_gpu_blocks_override=2) if test_preemption else dict(
gpu_memory_utilization=0.9))
dict(num_gpu_blocks_override=2) if test_preemption else dict(gpu_memory_utilization=0.9)
)
spec_mml = (spec_config or {}).get("max_model_len")
test_config = (f"executor={executor}, preemption={test_preemption}, "
f"async_sched={async_scheduling}, "
f"chunk_prefill={test_prefill_chunking}, "
f"spec_decoding={spec_decoding}, spec_mml={spec_mml}")
test_config = (
f"executor={executor}, preemption={test_preemption}, "
f"async_sched={async_scheduling}, "
f"chunk_prefill={test_prefill_chunking}, "
f"spec_decoding={spec_decoding}, spec_mml={spec_mml}"
)
print("-" * 80)
print(f"---- TESTING {test_str}: {test_config}")
print("-" * 80)
with VllmRunner(
model,
max_model_len=512,
enable_chunked_prefill=test_prefill_chunking,
# Force prefill chunking
max_num_batched_tokens=48 if test_prefill_chunking else None,
enforce_eager=True,
async_scheduling=async_scheduling,
distributed_executor_backend=executor,
dtype="float16", # avoid precision errors
speculative_config=spec_config,
disable_log_stats=False,
**cache_arg,
model,
max_model_len=512,
enable_chunked_prefill=test_prefill_chunking,
# Force prefill chunking
max_num_batched_tokens=48 if test_prefill_chunking else None,
enforce_eager=True,
async_scheduling=async_scheduling,
distributed_executor_backend=executor,
dtype="float16", # avoid precision errors
speculative_config=spec_config,
disable_log_stats=False,
**cache_arg,
) as vllm_model:
results = []
acceptance_rates: list[float] | None = [] if spec_decoding else None
@@ -197,26 +194,23 @@ def run_test(
results.append(
vllm_model.generate(
example_prompts,
sampling_params=SamplingParams(**default_params,
**override_params),
))
sampling_params=SamplingParams(**default_params, **override_params),
)
)
metrics_after = vllm_model.model.get_metrics()
if acceptance_rates is not None:
acceptance_rate = _get_acceptance_rate(metrics_before,
metrics_after)
acceptance_rate = _get_acceptance_rate(metrics_before, metrics_after)
acceptance_rates.append(acceptance_rate)
print(f"ACCEPTANCE RATE {acceptance_rate}")
if test_preemption:
preemptions = _get_count(metrics_before, metrics_after,
"vllm:num_preemptions")
preemptions = _get_count(metrics_before, metrics_after, "vllm:num_preemptions")
assert preemptions > 0, "preemption test had no preemptions"
if len(results) > 1:
# First check that the different parameter configs
# actually result in different output.
for other_test_outs, params in zip(results[1:],
sampling_param_tests[1:]):
for other_test_outs, params in zip(results[1:], sampling_param_tests[1:]):
with pytest.raises(AssertionError):
check_outputs_equal(
outputs_0_lst=results[0][0],