[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 5
parent
1085b64425
commit
651a88c09a
@@ -48,8 +48,7 @@ def test_without_spec_decoding(
|
||||
monkeypatch: pytest.MonkeyPatch,
|
||||
):
|
||||
"""Test consistency of combos of async scheduling, preemption,
|
||||
uni/multiproc executor, prefill chunking.
|
||||
"""
|
||||
uni/multiproc executor, prefill chunking."""
|
||||
struct_outputs = StructuredOutputsParams(json=sample_json_schema)
|
||||
test_sampling_params: list[dict[str, Any]] = [
|
||||
dict(),
|
||||
@@ -205,8 +204,7 @@ def run_tests(
|
||||
test_sampling_params: list[dict[str, Any]],
|
||||
):
|
||||
"""Test consistency of combos of async scheduling, preemption,
|
||||
uni/multiproc executor with spec decoding.
|
||||
"""
|
||||
uni/multiproc executor with spec decoding."""
|
||||
# Flex attention supports float32.
|
||||
attention_config = {"backend": "FLEX_ATTENTION"}
|
||||
|
||||
|
||||
@@ -72,8 +72,7 @@ def test_backend_per_kind_splits_groups(backend_per_kind, monkeypatch):
|
||||
)
|
||||
def test_backend_per_kind_overrides_global_backend(monkeypatch):
|
||||
"""A per-kind entry wins over the global ``backend`` for its kind; other
|
||||
kinds fall back to the global backend.
|
||||
"""
|
||||
kinds fall back to the global backend."""
|
||||
monkeypatch.setenv("VLLM_ALLOW_INSECURE_SERIALIZATION", "1")
|
||||
llm = LLM(
|
||||
model=MODEL,
|
||||
|
||||
@@ -34,8 +34,7 @@ def test_decoder_max_context_length_validation(
|
||||
max_tokens: int,
|
||||
) -> None:
|
||||
"""Check vLLM decoder model input validation for edge cases where
|
||||
the prompt length is (almost) equal to the max model length.
|
||||
"""
|
||||
the prompt length is (almost) equal to the max model length."""
|
||||
prompt_ids = [[43] * prompt_len]
|
||||
|
||||
with vllm_runner(
|
||||
@@ -72,8 +71,7 @@ def test_auto_fit_max_model_len_rejects_oversized_input(
|
||||
"""When max_model_len='auto' and KV cache memory is very limited,
|
||||
the engine auto-fits max_model_len to a small value. The frontend
|
||||
must see this reduced value and reject prompts that exceed it,
|
||||
rather than accepting them and hanging.
|
||||
"""
|
||||
rather than accepting them and hanging."""
|
||||
# Use a small KV cache budget to force auto-fit to a small
|
||||
# max_model_len. Pin block_size=16 so the budget is independent
|
||||
# of the platform's default block size. One block for this model is
|
||||
|
||||
@@ -172,8 +172,7 @@ def llm_v1():
|
||||
|
||||
def test_min_tokens_with_terminal_structured_output(llm_v1: LLM):
|
||||
"""A grammar whose only legal continuation is EOS must stop under
|
||||
min_tokens instead of sampling outside the grammar.
|
||||
"""
|
||||
min_tokens instead of sampling outside the grammar."""
|
||||
params = SamplingParams(
|
||||
temperature=GREEDY,
|
||||
min_tokens=20,
|
||||
|
||||
@@ -7,8 +7,7 @@ slightly different kernel/collective states (measured: same-state boots are
|
||||
bitwise identical across the two modes; cross-state boots differ by up to
|
||||
~0.25 logprob and can flip a near-tie token). A real sharding bug — rows
|
||||
routed to the wrong request, misaligned gathers — produces many-nats
|
||||
divergence and scrambled top-k sets, far outside these bounds.
|
||||
"""
|
||||
divergence and scrambled top-k sets, far outside these bounds."""
|
||||
|
||||
import pytest
|
||||
|
||||
@@ -107,8 +106,7 @@ def _assert_logprob_dicts_close(ref_lps, out_lps, what: str) -> None:
|
||||
@multi_gpu_test(num_gpus=2)
|
||||
def test_sharded_sampling_outputs_match(monkeypatch: pytest.MonkeyPatch):
|
||||
"""Generation, logprobs, and prompt logprobs match between batch-sharded
|
||||
sampling and the replicated fallback, up to measured boot-state noise.
|
||||
"""
|
||||
sampling and the replicated fallback, up to measured boot-state noise."""
|
||||
monkeypatch.setenv("VLLM_USE_V2_MODEL_RUNNER", "1")
|
||||
# Required for the collective_rpc mode check below.
|
||||
monkeypatch.setenv("VLLM_ALLOW_INSECURE_SERIALIZATION", "1")
|
||||
|
||||
Reference in New Issue
Block a user