iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Disable mypy arg-type and assignment checks in tests (#59428)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Harry Mellor
2026-09-30 15:18:50 +00:00
committed by GitHub
co-authored by Claude Opus 5.5
parent 17e9295dd5
commit fc2c801ad9
131 changed files with 402 additions and 409 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pyproject.toml
+5
View File
@@ -117,6 +117,11 @@ ignore_missing_imports = true
check_untyped_defs = true
follow_imports = "silent"
[[tool.mypy.overrides]]
module = "tests.*"
disable_error_code = ["arg-type", "assignment"]
enable_error_code = ["method-assign"]
[tool.pytest.ini_options]
markers = [
"slow_test",
tests/benchmarks/test_bfcl_dataset.py
+2 -2
View File
@@ -271,7 +271,7 @@ def test_bfcl_prompt_len_includes_tools(tmp_path: Path) -> None:
args.num_prompts = 1
with _patch_hf_api(fake_download):
samples = get_samples(args, fake) # type: ignore[arg-type]
samples = get_samples(args, fake)
assert len(samples) == 1
assert captured["tools"] is not None, (
@@ -311,7 +311,7 @@ def test_bfcl_prompt_len_falls_back_when_tokenizer_rejects_tools(
args.num_prompts = 1
with _patch_hf_api(fake_download):
samples = get_samples(args, _LegacyTokenizer()) # type: ignore[arg-type]
samples = get_samples(args, _LegacyTokenizer())
assert len(samples) == 1
assert samples[0].prompt_len > 0
tests/benchmarks/test_custom_dataset_chat_template_kwargs.py
+2 -2
View File
@@ -58,7 +58,7 @@ def test_chat_template_kwargs_forwarded(tmp_path: Path) -> None:
_write_one(jsonl)
tok = _RecordingTokenizer()
get_samples(_args(str(jsonl), {"thinking": True}), tok) # type: ignore[arg-type]
get_samples(_args(str(jsonl), {"thinking": True}), tok)
assert tok.captured_kwargs == {"thinking": True}
@@ -70,6 +70,6 @@ def test_chat_template_kwargs_default_is_noop(tmp_path: Path) -> None:
_write_one(jsonl)
tok = _RecordingTokenizer()
get_samples(_args(str(jsonl), None), tok) # type: ignore[arg-type]
get_samples(_args(str(jsonl), None), tok)
assert tok.captured_kwargs == {}
tests/compile/passes/distributed/test_async_tp.py
+1 -1
View File
@@ -285,7 +285,7 @@ def test_async_tp_pass_requires_full_graph_compilation():
]
async_tp_pass = object.__new__(AsyncTPPass)
async_tp_pass.compilation_config = vllm_config.compilation_config # type: ignore[assignment]
async_tp_pass.compilation_config = vllm_config.compilation_config
with pytest.raises(
AssertionError, match="AsyncTPPass requires full-graph compilation"
tests/compile/passes/distributed/test_fusion_all_reduce.py
+2 -2
View File
@@ -111,7 +111,7 @@ def test_fused_ar_workspace_hidden_dim(
target_hidden: int, draft_hidden: int | None, expected: int
):
config = _make_workspace_hidden_config(target_hidden, draft_hidden)
assert _fused_ar_workspace_hidden_dim(config) == expected # type: ignore[arg-type]
assert _fused_ar_workspace_hidden_dim(config) == expected
def test_fused_ar_workspace_hidden_dim_spec_without_draft():
@@ -120,7 +120,7 @@ def test_fused_ar_workspace_hidden_dim_spec_without_draft():
model_config=SimpleNamespace(get_hidden_size=lambda: 2048),
speculative_config=SimpleNamespace(draft_model_config=None),
)
assert _fused_ar_workspace_hidden_dim(config) == 2048 # type: ignore[arg-type]
assert _fused_ar_workspace_hidden_dim(config) == 2048
class TestAllReduceRMSNormModel(torch.nn.Module):
tests/compile/passes/distributed/test_sequence_parallelism.py
+1 -1
View File
@@ -222,7 +222,7 @@ def test_sequence_parallelism_pass_requires_full_graph_compilation():
]
sequence_parallelism_pass = object.__new__(SequenceParallelismPass)
sequence_parallelism_pass.compilation_config = vllm_config.compilation_config # type: ignore[assignment]
sequence_parallelism_pass.compilation_config = vllm_config.compilation_config
sequence_parallelism_pass.min_token_num = 1
with pytest.raises(
tests/compile/passes/test_pass_manager.py
+1 -1
View File
@@ -28,7 +28,7 @@ def test_bad_callable():
pass_manager.configure(config)
with pytest.raises(AssertionError):
pass_manager.add(simple_callable) # type: ignore[arg-type]
pass_manager.add(simple_callable)
# Pass that inherits from InductorPass
tests/compile/passes/test_silu_mul_quant_fusion.py
+1 -1
View File
@@ -331,7 +331,7 @@ def test_fusion_silu_and_mul_quant(
backend = TestBackend(*passes)
model = model_class(
hidden_size=hidden_size,
force_kernel=force_kernel, # type: ignore[arg-type]
force_kernel=force_kernel,
x=x,
dtype=dtype,
)
tests/config/test_config_generation.py
+2 -2
View File
@@ -133,7 +133,7 @@ def test_get_diff_sampling_param_includes_penalties():
},
)
diff = ModelConfig.get_diff_sampling_param(fake) # type: ignore[arg-type]
diff = ModelConfig.get_diff_sampling_param(fake)
assert diff["presence_penalty"] == 1.5
assert diff["frequency_penalty"] == 0.5
@@ -161,7 +161,7 @@ def test_generation_config_penalties_survive_diff_sampling_filter():
},
)
diff = ModelConfig.get_diff_sampling_param(fake) # type: ignore[arg-type]
diff = ModelConfig.get_diff_sampling_param(fake)
assert diff["presence_penalty"] == 1.1
assert diff["frequency_penalty"] == 0.7
tests/config/test_multimodal_config.py
+5 -5
View File
@@ -40,18 +40,18 @@ def test_supports_multimodal_inputs(model_id, limit_mm_per_prompt, expected):
def test_mm_encoder_attn_backend_str_conversion():
config = MultiModalConfig(mm_encoder_attn_backend="FLASH_ATTN") # type: ignore[arg-type]
config = MultiModalConfig(mm_encoder_attn_backend="FLASH_ATTN")
assert config.mm_encoder_attn_backend == AttentionBackendEnum.FLASH_ATTN
def test_mm_encoder_attn_backend_invalid():
with pytest.raises(ValueError):
MultiModalConfig(mm_encoder_attn_backend="not_a_backend") # type: ignore[arg-type]
MultiModalConfig(mm_encoder_attn_backend="not_a_backend")
def test_mm_hasher_algorithm_invalid():
with pytest.raises(ValueError, match="mm_hasher_algorithm"):
MultiModalConfig(mm_hasher_algorithm="md5") # type: ignore[arg-type]
MultiModalConfig(mm_hasher_algorithm="md5")
def test_mm_encoder_attn_backend_hash_updates():
@@ -560,7 +560,7 @@ def _resolve_mm_processor_device(
"""Run the `auto` resolution and report where the processor ended up."""
mm_config = MultiModalConfig(
mm_processor_kwargs={} if device is None else {"device": device},
mm_tensor_ipc=mm_tensor_ipc, # type: ignore[arg-type]
mm_tensor_ipc=mm_tensor_ipc,
)
model_config = MagicMock(spec=ModelConfig)
model_config.multimodal_config = mm_config
@@ -662,7 +662,7 @@ def _resolve_mm_video_decode_device(
the resulting video media IO kwargs."""
mm_config = MultiModalConfig(
mm_processor_kwargs={} if device is None else {"device": device},
mm_tensor_ipc=mm_tensor_ipc, # type: ignore[arg-type]
mm_tensor_ipc=mm_tensor_ipc,
media_io_kwargs={} if video_kwargs is None else {"video": dict(video_kwargs)},
)
model_config = MagicMock(spec=ModelConfig)
tests/conftest.py
+1 -1
View File
@@ -1163,7 +1163,7 @@ class VllmRunner:
output_logprobs = sample.logprobs
if include_prompt_token_ids:
outputs.append(
( # type: ignore[arg-type]
(
output_ids,
output_str,
output_logprobs,
tests/entrypoints/anthropic/test_anthropic_messages_conversion.py
+1 -1
View File
@@ -1447,7 +1447,7 @@ class TestDetectMergeInlineSystem:
model_config=None,
)
assert (
AnthropicServingMessages._should_merge_inline_system(online_renderer) # type: ignore[arg-type]
AnthropicServingMessages._should_merge_inline_system(online_renderer)
is expected
)
assert mock_logger.warning_once.called is expected
tests/entrypoints/cohere/test_cohere_chat_message.py
+2 -2
View File
@@ -29,7 +29,7 @@ from vllm.entrypoints.cohere.cohere_chat_message import (
class TestCitationSource:
def test_invalid_type_rejected(self):
with pytest.raises(ValidationError):
CitationSource(type="other") # type: ignore[arg-type]
CitationSource(type="other")
def test_none_fields_excluded_from_dump(self):
s = CitationSource(type="document", id="d1")
@@ -65,7 +65,7 @@ class TestCitationSource:
class TestCitation:
def test_invalid_type_rejected(self):
with pytest.raises(ValidationError):
Citation(type="OTHER") # type: ignore[arg-type]
Citation(type="OTHER")
def test_dump_excludes_none_fields(self):
c = Citation(start=0, end=5, text="hello")
tests/entrypoints/cohere/test_protocol.py
+2 -2
View File
@@ -136,7 +136,7 @@ class TestCohereChatV2Response:
with pytest.raises(ValidationError):
CohereChatV2Response(
id="r1",
finish_reason="NOT_A_REASON", # type: ignore[arg-type]
finish_reason="NOT_A_REASON",
message=msg,
)
@@ -246,5 +246,5 @@ class TestStreamingEventTypeField:
MessageStartEvent(
id="a",
delta={"message": {"role": "assistant"}},
type="other", # type: ignore[arg-type]
type="other",
)
tests/entrypoints/launchers/api_server/test_api_server_process_manager.py
+3 -3
View File
@@ -308,7 +308,7 @@ def test_external_process_monitoring(api_server_args):
try:
wait_for_completion_or_failure(
api_server_manager=manager,
coordinator=mock_coordinator, # type: ignore[arg-type]
coordinator=mock_coordinator,
)
except Exception as e:
result["exception"] = e
@@ -377,7 +377,7 @@ def test_gather_actual_addresses_end_to_end():
manager = APIServerProcessManager(
listen_address=f"tcp://{host}:0",
sock=sock,
args="test_args", # type: ignore[arg-type]
args="test_args",
num_servers=num_servers,
input_addresses=placeholder_inputs,
output_addresses=placeholder_outputs,
@@ -424,7 +424,7 @@ def test_gather_actual_addresses_child_crash_before_report():
manager = APIServerProcessManager(
listen_address=f"tcp://{host}:0",
sock=sock,
args="test_args", # type: ignore[arg-type]
args="test_args",
num_servers=num_servers,
input_addresses=placeholder_inputs,
output_addresses=placeholder_outputs,
tests/entrypoints/multimodal/llm/test_mm_processor_kwargs.py
+1 -1
View File
@@ -14,7 +14,7 @@ from vllm.outputs import RequestOutput
def _make_mock_llm() -> LLM:
llm = object.__new__(LLM)
llm.model_config = SimpleNamespace( # type: ignore[assignment]
llm.model_config = SimpleNamespace(
runner_type="generate", enable_prompt_embeds=False
)
return llm
tests/entrypoints/openai/test_render_parity.py
+1 -1
View File
@@ -225,7 +225,7 @@ def online_renderer(model_config: MockModelConfig, request) -> OnlineRenderer:
renderer.tokenizer = MagicMock()
return OnlineRenderer(
model_config=model_config, # type: ignore[arg-type]
model_config=model_config,
renderer=renderer,
request_logger=None,
chat_template=None,
tests/entrypoints/pooling/embed/test_io_processor.py
+1 -1
View File
@@ -411,7 +411,7 @@ class TestChunkedEmbeddingProcessing:
@classmethod
def _make_handler(cls):
handler = object.__new__(EmbedIOProcessor)
handler.model_config = cls._FakeModelConfig() # type: ignore[assignment]
handler.model_config = cls._FakeModelConfig()
handler.enable_chunked_processing = True
return handler
tests/entrypoints/pooling/scoring/test_io_processor_unit.py
+2 -2
View File
@@ -47,7 +47,7 @@ def llm_reranker_processor() -> CrossEncoderIOProcessor:
)
processor = CrossEncoderIOProcessor.__new__(CrossEncoderIOProcessor)
processor.model_config = SimpleNamespace(enable_prompt_embeds=False) # type: ignore[assignment]
processor.model_config = SimpleNamespace(enable_prompt_embeds=False)
processor.tokenizer = TokenizersBackend(
tokenizer_object=backend,
unk_token="[UNK]",
@@ -96,7 +96,7 @@ def test_token_type_ids_stay_aligned_with_a_truncated_padded_prompt():
)
processed_prompt = tok_params.apply_post_tokenization(
tokenizer, # type: ignore[arg-type]
tokenizer,
prompt,
)
assert "prompt_token_ids" in processed_prompt
tests/entrypoints/pooling/scoring/test_jina_ranking_io_processor_unit.py
+2 -4
View File
@@ -77,10 +77,8 @@ def offline_processor_and_context():
backend.pre_tokenizer = pre_tokenizers.WhitespaceSplit()
proc = JinaRankingIOProcessor.__new__(JinaRankingIOProcessor)
proc.tokenizer = TokenizersBackend(tokenizer_object=backend)
proc.model_config = SimpleNamespace( # type: ignore[assignment]
max_model_len=1024, is_encoder_decoder=False
)
proc.renderer = SimpleNamespace( # type: ignore[assignment]
proc.model_config = SimpleNamespace(max_model_len=1024, is_encoder_decoder=False)
proc.renderer = SimpleNamespace(
default_cmpl_tok_params=TokenizeParams(max_total_tokens=1024)
)
ctx = OfflineScoringInputsContext(
tests/entrypoints/pooling/test_io_processor.py
+2 -2
View File
@@ -53,8 +53,8 @@ def test_rejects_mismatched_lora_requests(processor: PoolingIOProcessor):
def test_rejects_conflicting_pooling_task(processor: PoolingIOProcessor):
processor.model_config = SimpleNamespace(is_encoder_decoder=False) # type: ignore[assignment]
processor.renderer = SimpleNamespace( # type: ignore[assignment] # Tokenization defaults only.
processor.model_config = SimpleNamespace(is_encoder_decoder=False)
processor.renderer = SimpleNamespace(
default_cmpl_tok_params=TokenizeParams(max_total_tokens=None)
)
ctx = OfflineEncodeInputsContext(
tests/entrypoints/serve/middleware/test_authentication_middleware.py
+1 -1
View File
@@ -87,7 +87,7 @@ def task_routes(request, monkeypatch) -> tuple[str, list[tuple[str, list[str]]]]
args,
app,
supported_tasks=(task,),
model_config=MockModelConfig(), # type: ignore[arg-type]
model_config=MockModelConfig(),
)
routes = get_all_http_routes(app)
tests/entrypoints/serve/tokenize/test_tokenization.py
+2 -2
View File
@@ -107,7 +107,7 @@ async def test_tokenize_chat(
conversation=conversation,
tokenize=False,
)
tokens = tokenizer.encode(prompt, add_special_tokens=add_special) # type: ignore[arg-type]
tokens = tokenizer.encode(prompt, add_special_tokens=add_special)
response = requests.post(
server.url_for("tokenize"),
@@ -176,7 +176,7 @@ async def test_tokenize_chat_with_tools(
tools=tools,
tokenize=False,
)
tokens = tokenizer.encode(prompt, add_special_tokens=add_special) # type: ignore[arg-type]
tokens = tokenizer.encode(prompt, add_special_tokens=add_special)
response = requests.post(
server.url_for("tokenize"),
tests/entrypoints/speech_to_text/transcription/test_qwen3_asr_sanitize_prompt.py
+1 -1
View File
@@ -47,7 +47,7 @@ def test_sanitize_handles_falsy_inputs() -> None:
assert _sanitize_transcription_user_text("") == ""
# The dataclass default for ``response_prefix`` is the empty string;
# the sanitizer must accept that without exception or extra work.
assert _sanitize_transcription_user_text(None) == "" # type: ignore[arg-type]
assert _sanitize_transcription_user_text(None) == ""
def test_sanitize_is_idempotent() -> None:
tests/entrypoints/unit_tests/test_chat_utils.py
+2 -2
View File
@@ -216,7 +216,7 @@ def video_url():
@pytest.fixture(scope="module")
def audio_url():
audio = AudioAsset("mary_had_lamb")
return encode_audio_url(*audio.audio_and_sample_rate) # type: ignore[arg-type]
return encode_audio_url(*audio.audio_and_sample_rate)
def _assert_mm_data_is_image_input(
@@ -3194,5 +3194,5 @@ def test_validate_chat_template_rejects_invalid_type():
with pytest.raises(
VLLMValidationError, match="not a valid chat template type"
) as exc_info:
validate_chat_template(123) # type: ignore[arg-type]
validate_chat_template(123)
assert exc_info.value.parameter == "chat_template"
tests/jit_monitor/test_hooks.py
+1 -1
View File
@@ -149,7 +149,7 @@ def test_activate_logs_info():
def test_activate_rejects_unknown_mode():
with pytest.raises(ValueError, match="Unsupported JIT monitor mode"):
jit_monitor.activate(mode="panic") # type: ignore[arg-type]
jit_monitor.activate(mode="panic")
def test_activate_without_triton():
tests/kernels/attention/test_attention_selector.py
+4 -4
View File
@@ -23,12 +23,12 @@ from vllm.platforms.interface import DeviceCapability
if current_platform.is_cuda():
from vllm.platforms.cuda import CudaPlatform
else:
CudaPlatform = None # type: ignore[assignment] # Unavailable platform import.
CudaPlatform = None
if current_platform.is_rocm():
from vllm.platforms.rocm import RocmPlatform
else:
RocmPlatform = None # type: ignore[misc, assignment] # Unavailable platform import.
RocmPlatform = None # type: ignore[misc] # Unavailable platform import.
from vllm.v1.attention.backend import AttentionType
from vllm.v1.attention.backends.registry import AttentionBackendEnum
@@ -356,14 +356,14 @@ def test_invalid_backend():
def test_auto_backend_string(auto_value: str):
"""Test that 'auto' string value triggers automatic backend selection."""
# Using "auto" should result in backend=None (automatic selection)
attention_config = AttentionConfig(backend=auto_value) # type: ignore[arg-type]
attention_config = AttentionConfig(backend=auto_value)
assert attention_config.backend is None
def test_auto_backend_selection_behavior():
"""Test that 'auto' backend behaves same as None (automatic selection)."""
# Create config with explicit "auto"
auto_config = AttentionConfig(backend="auto") # type: ignore[arg-type]
auto_config = AttentionConfig(backend="auto")
# Create config with None (default)
none_config = AttentionConfig(backend=None)
tests/kernels/attention/test_cpu_attn.py
+1 -1
View File
@@ -836,7 +836,7 @@ def varlen_encoder_zentorch_sdpa(
key,
value,
output,
SimpleNamespace(query_start_loc=query_start_loc, causal=False), # type: ignore[arg-type]
SimpleNamespace(query_start_loc=query_start_loc, causal=False),
scale,
sliding_window if sliding_window is not None else -1,
alibi_slopes,
tests/kernels/attention/test_minimax_m3.py
+1 -1
View File
@@ -1782,7 +1782,7 @@ def test_aiter_indexer_requires_the_aiter_attend(monkeypatch):
num_index_heads=1,
index_head_dim=HEAD_DIM,
# The implementation also accepts this legacy fp8 alias.
indexer_kv_dtype="fp8_e4m3", # type: ignore[arg-type]
indexer_kv_dtype="fp8_e4m3",
max_model_len=8192,
)
tests/kernels/attention/test_minimax_m3_msa_cutlass_sparse_decode.py
+5 -5
View File
@@ -235,14 +235,14 @@ def test_msa_metadata_builder_prepares_cutlass_for_regular_decode(
builder.num_q_heads = 64
builder.num_kv_heads = 4
builder.topk_blocks = TOPK
builder.kv_cache_spec = SimpleNamespace(num_kv_heads=4) # type: ignore[assignment]
builder.kv_cache_spec = SimpleNamespace(num_kv_heads=4)
builder.kv_cache_dtype = "fp8_e4m3"
builder.decode_backend = "cutlass"
builder.msa_cutlass_plan_cache = object() # type: ignore[assignment] # Mocked plan-cache sentinel.
builder.msa_cutlass_plan_cache = object()
metadata = builder.build(
0,
SimpleNamespace( # type: ignore[arg-type]
SimpleNamespace(
seq_lens_cpu_upper_bound=torch.full((batch,), 257, dtype=torch.int32)
),
)
@@ -377,7 +377,7 @@ def test_query_fp8_stays_valid_when_cutlass_plan_appears_on_replay(
with torch.cuda.stream(stream), override_forward_context(forward_context):
capture = BreakableCUDAGraphCapture()
with capture:
query_fp8 = MiniMaxM3SparseAttention._allocate_query_fp8(attention, qkv) # type: ignore[arg-type]
query_fp8 = MiniMaxM3SparseAttention._allocate_query_fp8(attention, qkv)
if query_fp8 is not None:
query_fp8.zero_()
run_attention(query_fp8)
@@ -387,7 +387,7 @@ def test_query_fp8_stays_valid_when_cutlass_plan_appears_on_replay(
assert capture.num_eager_breaks == 1
assert observed_ptrs == []
decode.msa_cutlass = object() # type: ignore[assignment]
decode.msa_cutlass = object()
for _ in range(3):
capture.replay()
stream.synchronize()
tests/kernels/attention/test_rocm_aiter_mla_causal_verify_mask.py
+1 -1
View File
@@ -204,7 +204,7 @@ def _run_verify_block():
),
):
assert isinstance(impl, AiterMLAImpl)
impl.forward_mqa((q_nope, q_pe), kv_cache, metadata, layer=None) # type: ignore[arg-type] # Spy decode path does not read layer.
impl.forward_mqa((q_nope, q_pe), kv_cache, metadata, layer=None)
return metadata, captured
tests/kernels/attention/test_rocm_aiter_mla_fp8_prefill.py
+2 -2
View File
@@ -121,7 +121,7 @@ def _build_prefill_metadata(
builder = object.__new__(AiterMLAMetadataBuilder)
builder.num_heads = num_heads
builder.mla_dims = SimpleNamespace(v_head_dim=V_HEAD_DIM) # type: ignore[assignment]
builder.mla_dims = SimpleNamespace(v_head_dim=V_HEAD_DIM)
builder._init_fp8_prefill_ps_buffers(
max_num_reqs=len(seq_lens),
max_prefill_qlen=max_q,
@@ -136,7 +136,7 @@ def _build_prefill_metadata(
prefill = SimpleNamespace(query_start_loc=qo_indptr, max_query_len=max_q)
metadata = SimpleNamespace(prefill=prefill, num_decodes=0)
common = SimpleNamespace(query_start_loc_cpu=qo_indptr_cpu)
builder._build_fp8_prefill_ps_metadata(metadata, common) # type: ignore[arg-type]
builder._build_fp8_prefill_ps_metadata(metadata, common)
return metadata, total_q, builder
tests/kernels/attention/test_rocm_aiter_mla_sink.py
+6 -6
View File
@@ -170,7 +170,7 @@ def test_sparse_mla_sink_matches_ragged_reference(
SimpleNamespace(_q_scale=q_scale, _k_scale=kv_scale),
q,
kv,
metadata, # type: ignore[arg-type]
metadata,
)
kv_flat = kv_ref[:, 0]
references = []
@@ -244,7 +244,7 @@ def test_sparse_mla_sink_rejects_unsupported_aiter_dtypes(
SimpleNamespace(_q_scale=None, _k_scale=None),
torch.empty(1, 16, Q_HEAD_DIM, dtype=q_dtype),
torch.empty(1, 1, Q_HEAD_DIM, dtype=kv_dtype),
metadata, # type: ignore[arg-type]
metadata,
)
@@ -384,7 +384,7 @@ def test_sparse_mla_sink_matches_dense_attention_with_empty_rows_and_paged_cache
paged_kv_indices=torch.tensor(indices, dtype=torch.int32, device="cuda"),
paged_kv_indptr=lengths.cumsum(0).to(device="cuda", dtype=torch.int32),
)
impl = ROCMAiterMLASparseImpl.__new__(ROCMAiterMLASparseImpl) # type: ignore[arg-type]
impl = ROCMAiterMLASparseImpl.__new__(ROCMAiterMLASparseImpl)
impl.num_heads = num_heads
impl.head_size = head_dim
impl.kv_lora_rank = value_dim
@@ -441,9 +441,9 @@ def test_sparse_mla_backend_resolves_only_contiguous_layer_layouts(monkeypatch,
]
if layout == "BLNHC":
with pytest.raises(ValueError, match="does not satisfy"):
resolve_kv_cache_layout(config, supported) # type: ignore[arg-type]
resolve_kv_cache_layout(config, supported)
else:
assert resolve_kv_cache_layout(config, supported).name == layout # type: ignore[arg-type]
assert resolve_kv_cache_layout(config, supported).name == layout
@pytest.mark.parametrize("dtype", [torch.bfloat16, torch.float16])
@@ -494,7 +494,7 @@ def test_sparse_mla_sink_forward_mqa_preserves_split_query(dtype):
actual, _ = impl.forward_mqa(
(q[..., :V_HEAD_DIM], q[..., V_HEAD_DIM:]),
kv,
metadata, # type: ignore[arg-type]
metadata,
SimpleNamespace(_q_scale=None, _k_scale=None),
)
references = [
tests/kernels/attention/test_rocm_aiter_mla_sparse_metadata_sync.py
+2 -2
View File
@@ -39,10 +39,10 @@ def _make_builder():
topk_tokens = 4
builder.device = torch.device("cpu")
builder.kv_cache_spec = SimpleNamespace(block_size=1) # type: ignore[assignment]
builder.kv_cache_spec = SimpleNamespace(block_size=1)
builder.model_dtype = torch.bfloat16
builder.kv_cache_dtype = "fp8"
builder.mla_dims = SimpleNamespace(kv_lora_rank=512, qk_rope_head_dim=64) # type: ignore[assignment]
builder.mla_dims = SimpleNamespace(kv_lora_rank=512, qk_rope_head_dim=64)
builder.topk_tokens = topk_tokens
builder.req_id_per_token_buffer = torch.zeros(
max_num_batched_tokens, dtype=torch.int32, device="cpu"
tests/kernels/attention/test_rocm_paged_mxfp4_indexer.py
+2 -4
View File
@@ -227,9 +227,7 @@ def _decode_metadata(case, rows, ratio, query_lens):
num_decode_tokens=len(rows),
num_prefills=0,
num_prefill_tokens=0,
decode=types.SimpleNamespace( # type: ignore[arg-type]
block_table=block_table, seq_lens=lens[:, None]
),
decode=types.SimpleNamespace(block_table=block_table, seq_lens=lens[:, None]),
decode_row_lens=lens,
decode_block_ends=(lens + CAND_BLOCK - 1) // CAND_BLOCK,
decode_native=native,
@@ -280,7 +278,7 @@ def _prefill_metadata(
num_decode_tokens=0,
num_prefills=len(case.seq_lens),
num_prefill_tokens=len(rows),
prefill=types.SimpleNamespace(chunks=chunks), # type: ignore[arg-type]
prefill=types.SimpleNamespace(chunks=chunks),
prefill_plans=plans,
gather_launches=plan_gather_launches(chunks, plans, gather_rows),
)
tests/kernels/attention/test_rocm_triton_attn_dsv4.py
+2 -2
View File
@@ -955,7 +955,7 @@ def test_rocm_capture_metadata_sets_adaptive_marker(monkeypatch) -> None:
)
builder = object.__new__(rocm_mod.DeepseekV4ROCMAiterMLASparseMetadataBuilder)
actual = builder.build_for_cudagraph_capture(SimpleNamespace()) # type: ignore[arg-type] # Patched base method ignores metadata.
actual = builder.build_for_cudagraph_capture(SimpleNamespace())
assert actual is metadata
assert actual.for_cudagraph_capture is _on_gfx950()
@@ -1563,7 +1563,7 @@ def test_dsv4_adaptive_mla_swa_metadata_graph_replay(monkeypatch) -> None:
"sliding_window": window_size,
},
)
vllm_config.speculative_config = SimpleNamespace( # type: ignore[assignment]
vllm_config.speculative_config = SimpleNamespace(
num_speculative_tokens=upper_query_len - 1,
parallel_drafting=False,
enable_adaptive_verification=True,
tests/kernels/mamba/test_gdn_forward_core_split.py
+1 -1
View File
@@ -176,7 +176,7 @@ def test_forward_core_split_matches_unified(
builder = GDNAttentionMetadataBuilder(
# GDNAttentionMetadataBuilder declares kv_cache_spec as AttentionSpec but
# asserts isinstance(kv_cache_spec, MambaSpec). MambaSpec is correct here.
kv_cache_spec=MambaSpec( # type: ignore[arg-type]
kv_cache_spec=MambaSpec(
block_size=BLOCK_SIZE, shapes=((16, 64),), dtypes=(torch.float16,)
),
layer_names=[PREFIX],
tests/kernels/moe/test_cutlass_moe.py
+1 -1
View File
@@ -538,7 +538,7 @@ def run_8_bit(
assert num_local_experts is not None
return run_with_expert_maps(
num_experts,
num_local_experts, # type: ignore[arg-type]
num_local_experts,
quant_config,
**kwargs,
)
tests/kernels/moe/test_ocp_mx_moe.py
+1 -1
View File
@@ -1724,7 +1724,7 @@ def test_rocm_mxfp4_moe_oracle(
w13_conv, w2_conv, w13_scale_conv, w2_scale_conv, w13_bias_conv, w2_bias_conv = (
convert_gpt_oss_weight_to_mxfp4_moe_kernel_format(
mxfp4_backend=backend,
layer=layer, # type: ignore[arg-type]
layer=layer,
w13_weight=w13_quant,
w2_weight=w2_quant,
w13_weight_scale=w13_scale,
tests/kernels/quantization/test_block_fp8.py
+1 -1
View File
@@ -259,7 +259,7 @@ def test_w8a8_block_fp8_torch_scaled_mm_matmul():
stub = BlockWiseTorchFP8ScaledMMLinearKernel.__new__(
BlockWiseTorchFP8ScaledMMLinearKernel
)
stub.config = types.SimpleNamespace(out_dtype=out_dtype) # type: ignore[assignment]
stub.config = types.SimpleNamespace(out_dtype=out_dtype)
out = stub.apply_block_scaled_mm(
A_fp8_cuda.cuda(), B_fp8.cuda(), As_cuda.cuda(), Bs.cuda()
)
tests/kernels/quantization/test_cutlass_w4a8.py
+1 -1
View File
@@ -208,7 +208,7 @@ def mm_test_helper(
a=tensors.a,
b_q=tensors.w_q,
b_group_scales=tensors.w_g_s,
b_group_size=group_size, # type: ignore[arg-type]
b_group_size=group_size,
b_channel_scales=tensors.w_ch_s,
a_token_scales=tensors.w_tok_s,
)
tests/kernels/quantization/test_marlin_tile_padding.py
+2 -4
View File
@@ -525,15 +525,13 @@ def test_check_marlin_supports_layer_allow_tile_padding():
)
# Tile-misaligned but group-aligned: rejected strictly, allowed w/ padding
layer: LinearBase = _FakeLinear( # type: ignore[assignment]
4640, 512, input_size=2048
)
layer: LinearBase = _FakeLinear(4640, 512, input_size=2048)
assert not check_marlin_supports_layer(layer, 128)
assert check_marlin_supports_layer(layer, 128, allow_tile_padding=True)
assert check_marlin_supports_layer(layer, -1, allow_tile_padding=True)
# A group straddling the TP shard cannot be fixed by padding
layer = _FakeLinear(4608, 4672, input_size=18688) # type: ignore[assignment]
layer = _FakeLinear(4608, 4672, input_size=18688)
assert not check_marlin_supports_layer(layer, 128, allow_tile_padding=True)
tests/kernels/quantization/test_nvfp4_emulation.py
+1 -1
View File
@@ -206,7 +206,7 @@ def test_nvfp4_emulation_support_check_rejects_bias_and_lora(
in_dtype=torch.bfloat16,
device="cuda",
routing_method=RoutingMethodType.TopK,
**config_kwargs, # type: ignore[arg-type]
**config_kwargs,
)
supported, reason = Nvfp4QuantizationEmulationTritonExperts.is_supported_config(
tests/model_executor/layers/test_fused_shared_expert.py
+3 -3
View File
@@ -413,7 +413,7 @@ def test_deepseek_v4_shared_expert_fse_uses_mtp_quantization_config_prefix(
)
compatible, reason = is_shared_expert_quant_fse_compatible(
DeepseekV4Config(), # type: ignore[arg-type]
DeepseekV4Config(),
"model.layers.2.ffn.experts",
"model.layers.2.ffn.shared_experts",
)
@@ -824,7 +824,7 @@ def test_models_fse_init(
model_type, quantization_config
)
vllm_config = VllmConfig()
vllm_config.model_config = SimpleNamespace( # type: ignore[assignment]
vllm_config.model_config = SimpleNamespace(
hf_config=config,
hf_text_config=config,
dtype=torch.bfloat16,
@@ -893,7 +893,7 @@ def test_models_fse_init(
DeepSeekV4MTP,
)
vllm_config.speculative_config = SimpleNamespace( # type: ignore[assignment]
vllm_config.speculative_config = SimpleNamespace(
draft_model_config=SimpleNamespace(hf_config=config),
method="mtp",
parallel_drafting=False,
tests/model_executor/layers/test_mla_short_prefill_indexer.py
+4 -4
View File
@@ -46,7 +46,7 @@ def test_sparse_attention_refreshes_batch_state_inside_eager_segment(
with pytest.raises(BatchStateRefreshed):
DeepseekV32Attention._sparse_indexer_and_attn(
layer, # type: ignore[arg-type]
layer,
torch.empty(1, dtype=torch.long),
torch.empty(1, 1),
torch.empty(1, 1, 1),
@@ -80,7 +80,7 @@ def make_indexer_metadata(
num_decode_tokens=num_decode_tokens,
num_prefills=num_prefills,
num_prefill_tokens=num_prefill_tokens,
prefill=SimpleNamespace(chunks=[]) if num_prefills else None, # type: ignore[arg-type]
prefill=SimpleNamespace(chunks=[]) if num_prefills else None,
)
@@ -129,7 +129,7 @@ def test_short_prefill_updates_k_cache_before_scoring_decision(
slot_mapping=slot_mapping,
)
if indexer_metadata.num_decodes:
indexer_metadata.decode = object() # type: ignore[assignment]
indexer_metadata.decode = object()
mla_metadata = make_mla_metadata(
use_dense_mha=batch_kind != "force_mqa",
num_decode_tokens=mla_num_decode_tokens,
@@ -307,7 +307,7 @@ def test_deepseek_v32_dispatches_selected_mha(
output = torch.empty(2, 2)
DeepseekV32Attention._sparse_indexer_and_attn(
layer, # type: ignore[arg-type]
layer,
torch.arange(2),
torch.empty(2, 2),
q_nope,
tests/model_executor/layers/test_pooler_activations.py
+2 -2
View File
@@ -224,11 +224,11 @@ class TestResolveClassifierActFn:
model_config = SimpleNamespace(
hf_config=SimpleNamespace(num_labels=3, problem_type="")
)
result = resolve_classifier_act_fn(model_config, act_fn=None) # type: ignore[arg-type]
result = resolve_classifier_act_fn(model_config, act_fn=None)
assert isinstance(result, PoolerClassify)
assert result.num_labels == 3
def test_passes_through_provided_act_fn(self):
custom = PoolerIdentity()
result = resolve_classifier_act_fn(None, act_fn=custom) # type: ignore[arg-type]
result = resolve_classifier_act_fn(None, act_fn=custom)
assert result is custom
tests/model_executor/layers/test_pooler_methods.py
+1 -1
View File
@@ -320,7 +320,7 @@ def test_dispatch_seq_cls_honors_token_pooling_type(tok_pooling_type):
),
)
classifier = torch.nn.Linear(4, 3)
with set_current_vllm_config(config): # type: ignore[arg-type]
with set_current_vllm_config(config):
pooler = DispatchPooler.for_seq_cls(pooler_config, classifier=classifier)
token_ids = [[2, 99, 3, 99], [4, 99], [5, 6]]
tests/model_executor/model_loader/tensorizer_loader/test_tensorizer.py
+1 -1
View File
@@ -36,7 +36,7 @@ try:
import tensorizer
from tensorizer import EncryptionParams
except ImportError:
tensorizer = PlaceholderModule("tensorizer") # type: ignore[assignment]
tensorizer = PlaceholderModule("tensorizer")
EncryptionParams = tensorizer.placeholder_attr("EncryptionParams")
tests/model_executor/model_loader/test_weight_cache.py
+1 -1
View File
@@ -368,4 +368,4 @@ def test_ipc_loader_copy_mode_reports_no_external_weight_memory():
load_format="ipc_cache", model_loader_extra_config={"mode": "copy"}
)
loader = IpcModelLoader(copy_mode)
assert loader.get_external_weight_memory(None) == 0 # type: ignore[arg-type]
assert loader.get_external_weight_memory(None) == 0
tests/models/language/generation/test_mistral.py
+3 -3
View File
@@ -259,7 +259,7 @@ def test_mistral_function_calling(vllm_runner, model: str, dtype: str) -> None:
model_output = outputs[0].outputs[0].text.strip()
assert model_output.startswith(tool_parser.bot_token), model_output
parsed_message = tool_parser.extract_tool_calls(model_output, None) # type: ignore[arg-type]
parsed_message = tool_parser.extract_tool_calls(model_output, None)
assert parsed_message.tools_called
@@ -302,7 +302,7 @@ def test_mistral_function_call_nested_json():
model_output = f"{parser.bot_token}get_current_weather{json.dumps(args_dict)}"
parsed = parser.extract_tool_calls(model_output, None) # type: ignore[arg-type]
parsed = parser.extract_tool_calls(model_output, None)
# Assertions: the tool call is detected and the full nested JSON is parsed
# without truncation.
@@ -335,7 +335,7 @@ def test_mistral_function_call_nested_json():
]
)
parsed = parser.extract_tool_calls(model_output, None) # type: ignore[arg-type]
parsed = parser.extract_tool_calls(model_output, None)
# Assertions: the tool call is detected and the full nested JSON is parsed
# without truncation.
tests/models/multimodal/generation/vlm_utils/model_utils.py
+1 -1
View File
@@ -1193,7 +1193,7 @@ def voxtral_patch_hf_runner(hf_model: "HfRunner") -> "HfRunner":
output.sequences = output.sequences[:, prompt_len:]
return output
hf_model.get_inputs = patched_get_inputs # type: ignore[method-assign, assignment]
hf_model.get_inputs = patched_get_inputs # type: ignore[method-assign]
hf_model.model.generate = patched_generate # type: ignore[method-assign]
return hf_model
tests/models/multimodal/pooling/test_clip.py
+1 -1
View File
@@ -132,7 +132,7 @@ def test_models(
_run_test(
hf_runner,
vllm_runner,
input_cases, # type: ignore[arg-type]
input_cases,
model,
dtype=dtype,
)
tests/models/multimodal/pooling/test_llama_nemotron_vl.py
+1 -1
View File
@@ -190,7 +190,7 @@ def test_models(
_run_test(
hf_runner,
vllm_runner,
input_cases, # type: ignore[arg-type]
input_cases,
model,
dtype=dtype,
)
tests/models/multimodal/pooling/test_siglip.py
+1 -1
View File
@@ -149,7 +149,7 @@ def test_models(
_run_test(
hf_runner,
vllm_runner,
input_cases, # type: ignore[arg-type]
input_cases,
model,
dtype=dtype,
)
tests/models/multimodal/processing/test_qwen2_vl.py
+2 -2
View File
@@ -106,7 +106,7 @@ def test_merge_and_resolve_mm_processor_kwargs_preserves_request_alias_precedenc
mm_device_do_normalize=False,
)
model_config = SimpleNamespace(get_multimodal_config=lambda: mm_config)
ctx = InputProcessingContext(model_config, tokenizer=None) # type: ignore[arg-type]
ctx = InputProcessingContext(model_config, tokenizer=None)
info = Qwen2VLProcessingInfo(ctx)
monkeypatch.setattr(
info,
@@ -236,7 +236,7 @@ def test_get_vision_info_reads_merged_modality_scope(
do_resize=False,
image_processor=image_processor,
mm_kwargs={},
modality=modality, # type: ignore[arg-type]
modality=modality,
)
assert seen == [merged[scope]]
tests/models/transformers/test_backend.py
+2 -2
View File
@@ -260,7 +260,7 @@ def test_quantization(
model,
model_impl="auto",
enforce_eager=True,
**quantization_kwargs, # type: ignore[arg-type]
**quantization_kwargs,
) as vllm_model:
vllm_outputs = vllm_model.generate_greedy_logprobs(
example_prompts, max_tokens=max_tokens, num_logprobs=num_logprobs
@@ -270,7 +270,7 @@ def test_quantization(
model,
model_impl="transformers",
enforce_eager=True,
**quantization_kwargs, # type: ignore[arg-type]
**quantization_kwargs,
) as vllm_model:
model_config = vllm_model.llm.llm_engine.model_config
assert model_config.using_transformers_backend()
tests/multimodal/media/test_video.py
+1 -1
View File
@@ -385,7 +385,7 @@ def test_pynvvideocodec_unrelated_error_propagates(
PyNvVideoCodecVideoBackendMixin.decode_frames_pynvvideocodec(
None,
b"video",
None, # type: ignore[arg-type] # Fails before target metadata is used.
None,
)
assert exc_info.value is original_error
tests/multimodal/test_audio.py
+1 -1
View File
@@ -179,7 +179,7 @@ def test_audio_resampler_invalid_method(dummy_audio):
# Validated eagerly so a bad method fails at construction, not on the
# first audio request.
with pytest.raises(ValueError, match="Invalid resampling method"):
AudioResampler(target_sr=22050, method="invalid") # type: ignore[arg-type]
AudioResampler(target_sr=22050, method="invalid")
def test_audio_resampler_no_target_sr(dummy_audio):
tests/multimodal/test_cache.py
+12 -12
View File
@@ -276,14 +276,14 @@ def test_oversized_item_is_served_uncached():
item = MultiModalKwargsItem.dummy(nbytes=4096)
small = MultiModalKwargsItem.dummy(nbytes=64)
p0_only = MultiModalProcessorOnlyCache(model_config) # type: ignore[arg-type]
p0_only = MultiModalProcessorOnlyCache(model_config)
assert p0_only.get_and_update_item((item, []), "big")[0] is item
assert not p0_only.is_cached_item("big")
assert p0_only.get_and_update_item((small, []), "small")[0] is small
assert p0_only.is_cached_item("small")
p0 = LruKeyReplicatedSenderCache(model_config) # type: ignore[arg-type]
p1 = LruKeyReplicatedReceiverCache(model_config) # type: ignore[arg-type]
p0 = LruKeyReplicatedSenderCache(model_config)
p1 = LruKeyReplicatedReceiverCache(model_config)
assert item is not None
assert p0.get_and_update_item((item, []), "big")[0] is item
assert not p0.is_cached_item("big")
@@ -304,8 +304,8 @@ def test_mm_cache_miss_raises_and_recovers():
model or network.
"""
model_config = _StubModelConfig(mm_processor_cache_gb=1)
p0 = LruKeyReplicatedSenderCache(model_config) # type: ignore[arg-type]
p1 = LruKeyReplicatedReceiverCache(model_config) # type: ignore[arg-type]
p0 = LruKeyReplicatedSenderCache(model_config)
p1 = LruKeyReplicatedReceiverCache(model_config)
mm_hash = "image_A"
item = MultiModalKwargsItem.dummy(nbytes=64)
@@ -348,7 +348,7 @@ def test_receiver_cache_replaces_stale_item_when_payload_resent():
and kills EngineCore. Prefer the fresh payload and keep it for later hits.
"""
model_config = _StubModelConfig(mm_processor_cache_gb=1)
p1 = LruKeyReplicatedReceiverCache(model_config) # type: ignore[arg-type]
p1 = LruKeyReplicatedReceiverCache(model_config)
small = MultiModalKwargsItem.dummy(nbytes=64)
large = MultiModalKwargsItem.dummy(nbytes=256)
mm_hash = "shared-id"
@@ -361,7 +361,7 @@ def test_receiver_cache_replaces_stale_item_when_payload_resent():
def test_receiver_cache_features_keep_resent_payload():
"""EngineCore updates features through get_and_update_features."""
model_config = _StubModelConfig(mm_processor_cache_gb=1)
p1 = LruKeyReplicatedReceiverCache(model_config) # type: ignore[arg-type]
p1 = LruKeyReplicatedReceiverCache(model_config)
small = MultiModalKwargsItem.dummy(nbytes=64)
large = MultiModalKwargsItem.dummy(nbytes=256)
mm_hash = "shared-id"
@@ -394,7 +394,7 @@ def test_mm_cache_miss_batches_all_drifted_hashes():
non-drifted items in the same request are still ingested.
"""
model_config = _StubModelConfig(mm_processor_cache_gb=1)
p1 = LruKeyReplicatedReceiverCache(model_config) # type: ignore[arg-type]
p1 = LruKeyReplicatedReceiverCache(model_config)
item = MultiModalKwargsItem.dummy(nbytes=64)
def _feature(
@@ -442,8 +442,8 @@ def test_shm_receiver_handles_prefix_covered_items(monkeypatch):
parallel_config=SimpleNamespace(world_size=1),
model_config=_StubModelConfig(mm_processor_cache_gb=4 * MiB_bytes / GiB_bytes),
)
p0 = ShmObjectStoreSenderCache(vllm_config) # type: ignore[arg-type]
p1 = ShmObjectStoreReceiverCache(vllm_config, mp.Lock()) # type: ignore[arg-type]
p0 = ShmObjectStoreSenderCache(vllm_config)
p1 = ShmObjectStoreReceiverCache(vllm_config, mp.Lock())
def _feature(
mm_hash: str, data: MultiModalKwargsItem | None
@@ -1082,9 +1082,9 @@ async def test_release_kv_cache_resends_mm_payload(use_async, release_error):
async def call_release():
if use_async:
await AsyncLLM.release_kv_cache_memory(engine) # type: ignore[arg-type]
await AsyncLLM.release_kv_cache_memory(engine)
else:
LLMEngine.release_kv_cache_memory(engine) # type: ignore[arg-type]
LLMEngine.release_kv_cache_memory(engine)
if release_error:
with pytest.raises(RuntimeError, match=release_error):
tests/multimodal/test_processing.py
+4 -4
View File
@@ -1281,7 +1281,7 @@ def test_get_supported_mm_processor_kwargs_uses_supported_modalities(
processor: SimpleNamespace,
expected: dict[str, set[str]],
) -> None:
info = BaseProcessingInfo(SimpleNamespace()) # type: ignore[arg-type]
info = BaseProcessingInfo(SimpleNamespace())
info.__dict__["supported_mm_limits"] = supported_mm_limits
monkeypatch.setattr(info, "get_hf_processor", lambda **_: processor)
@@ -1291,7 +1291,7 @@ def test_get_supported_mm_processor_kwargs_uses_supported_modalities(
def test_supported_mm_processor_kwargs_is_cached(
monkeypatch: pytest.MonkeyPatch,
) -> None:
info = BaseProcessingInfo(SimpleNamespace()) # type: ignore[arg-type]
info = BaseProcessingInfo(SimpleNamespace())
expected = {"images_kwargs": {"size"}}
calls = 0
@@ -1442,7 +1442,7 @@ def test_get_merged_mm_kwargs_treats_empty_scopes_as_absent_before_routing(
mm_device_do_normalize=False,
)
model_config = SimpleNamespace(get_multimodal_config=lambda: mm_config)
ctx = InputProcessingContext(model_config, tokenizer=None) # type: ignore[arg-type]
ctx = InputProcessingContext(model_config, tokenizer=None)
merged = ctx.get_merged_mm_kwargs(
inference_kwargs,
@@ -1482,7 +1482,7 @@ def test_get_merged_mm_kwargs_merges_before_routing():
mm_device_do_normalize=False,
)
model_config = SimpleNamespace(get_multimodal_config=lambda: mm_config)
ctx = InputProcessingContext(model_config, tokenizer=None) # type: ignore[arg-type]
ctx = InputProcessingContext(model_config, tokenizer=None)
merged = ctx.get_merged_mm_kwargs(
{
tests/multimodal/test_registry.py
+1 -1
View File
@@ -24,4 +24,4 @@ def test_create_processor_error_uses_served_model_name():
ValueError,
match="friendly-model-name is not a multimodal model",
):
MULTIMODAL_REGISTRY.create_processor(model_config) # type: ignore[arg-type]
MULTIMODAL_REGISTRY.create_processor(model_config)
tests/multimodal/test_video.py
+1 -1
View File
@@ -493,7 +493,7 @@ def test_pynvvideocodec_failed_rebuild_invalidates_decoder_slot():
old_decoder = FakeDecoder()
slot = PyNvVideoCodecDecoderSlot(FakeStream())
slot.decoder = old_decoder # type: ignore[assignment] # Fake retained GPU decoder.
slot.decoder = old_decoder
slot.source_path = "valid.mp4"
class FakeNvc:
tests/parser/cohere/test_structural_tags.py
+1 -1
View File
@@ -347,7 +347,7 @@ class TestAdjustRequestFoldFromStructuredOutputs:
id="chat_completion_request",
),
pytest.param(
lambda: StructuredOutputsParams(json=[1, 2, 3]), # type: ignore[arg-type]
lambda: StructuredOutputsParams(json=[1, 2, 3]),
id="structured_outputs_params",
),
],
tests/plugins_tests/bitsandbytes/test_bitsandbytes.py
+1 -1
View File
@@ -271,7 +271,7 @@ def test_bitsandbytes_passes_revision_by_name():
return_value=["/folder/model.safetensors"],
),
):
bnb.BitsAndBytesModelLoader._prepare_weights(fake_self, "org/model", "myrev") # type: ignore[arg-type]
bnb.BitsAndBytesModelLoader._prepare_weights(fake_self, "org/model", "myrev")
mock_idx.assert_called_once()
assert mock_idx.call_args.kwargs.get("revision") == "myrev"
tests/plugins_tests/bitsandbytes/test_transformers.py
+2 -2
View File
@@ -43,7 +43,7 @@ def test_transformers_backend_quantization(
model,
model_impl="auto",
enforce_eager=True,
**quantization_kwargs, # type: ignore[arg-type]
**quantization_kwargs,
) as vllm_model:
vllm_outputs = vllm_model.generate_greedy_logprobs(
example_prompts, max_tokens=max_tokens, num_logprobs=num_logprobs
@@ -53,7 +53,7 @@ def test_transformers_backend_quantization(
model,
model_impl="transformers",
enforce_eager=True,
**quantization_kwargs, # type: ignore[arg-type]
**quantization_kwargs,
) as vllm_model:
model_config = vllm_model.llm.llm_engine.model_config
assert model_config.using_transformers_backend()
tests/plugins_tests/gguf/test_gguf_plugin_multimodal.py
+1 -1
View File
@@ -92,7 +92,7 @@ def run_multimodal_gguf_test(
num_logprobs: int,
):
# Load images at runtime (inside subprocess) to avoid pickle issues
images = [ImageAsset(name).pil_image for name in model.image_names] # type: ignore[arg-type]
images = [ImageAsset(name).pil_image for name in model.image_names]
size_factors = [0.25, 0.5, 1.0]
inputs_per_image = [
(
tests/plugins_tests/test_endpoint_plugins.py
+2 -2
View File
@@ -172,7 +172,7 @@ async def test_init_state_is_noop_without_phase_a(monkeypatch: pytest.MonkeyPatc
monkeypatch.setenv("VLLM_PLUGINS", "dummy_admin_endpoint_plugin")
state = State()
await init_endpoint_plugins_state(_FakeEngineClient(), state, _build_args()) # type: ignore[arg-type]
await init_endpoint_plugins_state(_FakeEngineClient(), state, _build_args())
assert not hasattr(state, "dummy_engine_client")
@@ -223,7 +223,7 @@ async def test_endpoint_plugin_end_to_end(monkeypatch: pytest.MonkeyPatch):
)
fake_engine_client = _FakeEngineClient(rpc_result=["cfg-a", "cfg-b"])
await init_endpoint_plugins_state(fake_engine_client, app.state, args) # type: ignore[arg-type]
await init_endpoint_plugins_state(fake_engine_client, app.state, args)
assert app.state.dummy_engine_client is fake_engine_client
tests/plugins_tests/test_oot_registration_offline.py
+1 -1
View File
@@ -92,7 +92,7 @@ def test_oot_registration_multimodal(
)
first_token = llm.get_tokenizer().decode(0)
outputs = llm.generate(prompts, sampling_params) # type: ignore[arg-type]
outputs = llm.generate(prompts, sampling_params)
for output in outputs:
generated_text = output.outputs[0].text
tests/plugins_tests/test_stats_logger_plugins.py
+1 -1
View File
@@ -24,7 +24,7 @@ def test_stat_logger_plugin_is_discovered(monkeypatch: pytest.MonkeyPatch):
vllm_config = VllmConfig()
instance = factories[0](vllm_config)
# The `is` check above narrows DummyStatLogger to a Callable
assert isinstance(instance, DummyStatLogger) # type: ignore[arg-type]
assert isinstance(instance, DummyStatLogger)
def test_no_plugins_loaded_if_env_empty(monkeypatch: pytest.MonkeyPatch):
tests/quantization/test_auto_gptq.py
+2 -2
View File
@@ -123,8 +123,8 @@ def test_auto_gptq_moe_creates_zero_initialized_expert_biases():
method = object.__new__(AutoGPTQMoEMethod)
method.quant_config = AutoGPTQConfig(4, 128, False, True, False, {}, {})
method.input_dtype = None
method.experts_cls = None # type: ignore[assignment] # Weight creation does not select experts.
method.moe = SimpleNamespace(w13_num_shards=2) # type: ignore[assignment]
method.experts_cls = None
method.moe = SimpleNamespace(w13_num_shards=2)
layer = torch.nn.Module()
method.create_weights(
tests/quantization/test_auto_round.py
+6 -6
View File
@@ -423,7 +423,7 @@ def test_inc_get_quant_method_unquantized_routed_experts_with_model_prefix(
config = make_config(extra_config={"model.moe_layer": {"bits": 16}})
layer = object.__new__(RoutedExperts)
layer.moe_config = None # type: ignore[assignment] # No MoE config used by this stub.
layer.moe_config = None
method = config.get_quant_method(layer, "moe_layer")
@@ -730,7 +730,7 @@ def test_wna16_cuda_low_bit_moe_routes_to_humming(monkeypatch, bits) -> None:
)
layer = object.__new__(RoutedExperts)
layer.moe_config = DummyMoeConfig() # type: ignore[assignment]
layer.moe_config = DummyMoeConfig()
layer_config = make_layer_config(bits=bits)
method = INCWna16Scheme().get_moe_method(
make_config(), layer, "model.layers.0.mlp", layer_config
@@ -861,7 +861,7 @@ def test_qwen3_30b_a3b_w4a16_autoround_routes_to_gptq_moe(
config = make_qwen3_autoround_config("qwen3_30b_a3b_w4a16")
layer = object.__new__(RoutedExperts)
layer.moe_config = DummyMoeConfig() # type: ignore[assignment]
layer.moe_config = DummyMoeConfig()
method = config.get_quant_method(layer, "model.layers.0.mlp")
@@ -890,7 +890,7 @@ def test_qwen3_30b_a3b_mxfp4_autoround_routes_to_mxfp4_moe(
config = make_qwen3_autoround_config("qwen3_30b_a3b_mxfp4")
layer = object.__new__(RoutedExperts)
layer.moe_config = DummyMoeConfig() # type: ignore[assignment]
layer.moe_config = DummyMoeConfig()
ignored_method = config.get_quant_method(
object.__new__(LinearBase), "model.layers.0.self_attn.q_proj"
@@ -1186,7 +1186,7 @@ def test_wna16_xpu_moe_routes_to_gptq_moe(monkeypatch) -> None:
)
layer = object.__new__(RoutedExperts)
layer.moe_config = DummyMoeConfig() # type: ignore[assignment]
layer.moe_config = DummyMoeConfig()
method = INCWna16Scheme().get_moe_method(
make_config(),
layer,
@@ -1669,7 +1669,7 @@ def test_inc_get_quant_method_unquantized_moe_returns_unquantized(
when extra_config has bits >= 16."""
config = make_config(extra_config={"layer": {"bits": 16}})
layer = object.__new__(RoutedExperts)
layer.moe_config = None # type: ignore[assignment] # No MoE config used by this stub.
layer.moe_config = None
class DummyUnquantizedFusedMoEMethod:
def __init__(self, moe_config) -> None:
tests/quantization/test_fp8.py
+2 -2
View File
@@ -220,7 +220,7 @@ def test_deepseek_v41_vl_mapper_routes_linear_scales(
quant_config=SimpleNamespace(weight_block_size=weight_block_size)
)
resolved = model_module._linear_scale_param_name(
vllm_config, # type: ignore[arg-type]
vllm_config,
expert_dtype,
)
assert resolved == scale_name
@@ -798,7 +798,7 @@ def test_kv_cache_scale_sync_to_host_copies():
set_default_quant_scales(layer, register_buffer=True)
layer.kv_cache_dtype = "fp8"
method = BaseKVCacheMethod(quant_config=None) # type: ignore[arg-type] # Scale synchronization does not use quant_config.
method = BaseKVCacheMethod(quant_config=None)
method.create_weights(layer)
# 0.3 stays != 1.0 even after the fp8_fnuz x2 rescale.
checkpoint_scale = torch.tensor(0.3, dtype=torch.float32)
tests/quantization/test_moe_wna16.py
+4 -4
View File
@@ -218,9 +218,9 @@ def test_gptq_nfirst_weights_pass_through_for_triton():
def test_moe_wna16_setup_forwards_selected_backend(monkeypatch):
method = object.__new__(MoeWNA16Method)
method.experts_cls = object # type: ignore[assignment] # Opaque mocked-expert sentinel.
method.experts_cls = object
method.wna16_backend = WNA16MoEBackend.HUMMING
method.moe = object() # type: ignore[assignment] # Opaque mocked-config sentinel.
method.moe = object()
quant_config = object()
monkeypatch.setattr(
method, "get_fused_moe_quant_config", lambda layer: quant_config
@@ -235,7 +235,7 @@ def test_moe_wna16_setup_forwards_selected_backend(monkeypatch):
monkeypatch.setattr(moe_wna16, "make_wna16_moe_kernel", fake_make_wna16_moe_kernel)
method._setup_kernel(layer) # type: ignore[arg-type] # Routing-table stub.
method._setup_kernel(layer)
assert method.moe_kernel is kernel
assert captured["backend"] == WNA16MoEBackend.HUMMING
@@ -280,7 +280,7 @@ def test_moe_wna16_uses_humming_quant_config(monkeypatch):
),
)
assert method.get_fused_moe_quant_config(layer) is quant_config # type: ignore[arg-type] # Identity sentinel.
assert method.get_fused_moe_quant_config(layer) is quant_config
@pytest.mark.skipif(
tests/quantization/test_online.py
+4 -4
View File
@@ -210,9 +210,9 @@ def test_online_nvfp4_reuses_kernel_when_weights_are_reprocessed(
monkeypatch,
) -> None:
method = object.__new__(Nvfp4OnlineMoEMethod)
method.moe = SimpleNamespace(is_act_and_mul=True) # type: ignore[assignment]
method.nvfp4_backend = object() # type: ignore[assignment] # Opaque mocked-backend sentinel.
method.experts_cls = object # type: ignore[assignment] # Opaque mocked-expert sentinel.
method.moe = SimpleNamespace(is_act_and_mul=True)
method.nvfp4_backend = object()
method.experts_cls = object
method.moe_quant_config = None
method.moe_kernel = None
@@ -1432,7 +1432,7 @@ def test_online_int8_moe_w2_scale_matches_unsharded(monkeypatch) -> None:
layer.w2_weight = torch.nn.Parameter(w2_in, requires_grad=False)
layer.num_experts = layer.local_num_experts = w13.shape[0]
method = SimpleNamespace(moe=SimpleNamespace(tp_size=moe_tp_size))
Int8OnlineMoEMethod._quantize_weights(method, layer) # type: ignore[arg-type]
Int8OnlineMoEMethod._quantize_weights(method, layer)
return layer.w2_weight, layer.w2_scale
full_weight, full_scale = quantize(w2, 1)
tests/quantization/test_quark.py
+2 -2
View File
@@ -2314,7 +2314,7 @@ class TestQuarkInt4Format:
None,
quant_config.quant_config["global_quant_config"]["weight"],
quant_config.pack_method,
moe_config, # type: ignore[arg-type]
moe_config,
)
layer = _FakeLayer(moe_config)
@@ -2326,7 +2326,7 @@ class TestQuarkInt4Format:
requires_grad=False,
)
loader = method.get_weight_loader(
layer, # type: ignore[arg-type]
layer,
weight_loader=None,
)
loader(
tests/quantization/test_trtllm_nvfp4_hidden_dim_padding.py
+1 -1
View File
@@ -34,7 +34,7 @@ def test_shared_nvfp4_input_scales_have_writable_storage(monkeypatch):
outputs = prepare_nvfp4_moe_layer_for_fi_or_cutlass(
backend=NvFp4MoeBackend.FLASHINFER_CUTLASS,
layer=layer, # type: ignore[arg-type]
layer=layer,
w13=w13,
w13_scale=w13_scale,
w13_scale_2=weight_scale,
tests/quantization/test_turboquant.py
+1 -1
View File
@@ -300,7 +300,7 @@ class TestTurboQuantKVCacheSpec:
# The layer builds an unpacked spec; the worker's spec-collection
# loop applies TQ slot packing via the backend's customize_spec hook.
spec = Attention.get_kv_cache_spec(layer, vllm_config) # type: ignore[arg-type]
spec = Attention.get_kv_cache_spec(layer, vllm_config)
assert isinstance(spec, FullAttentionSpec)
assert spec.kv_quant_mode.is_turboquant
assert spec.state_content_bytes is None
tests/reasoning/test_base_thinking_reasoning_parser.py
+1 -1
View File
@@ -68,7 +68,7 @@ class TestBaseThinkingReasoningParserInit:
def test_initialization_with_missing_tokenizer(self):
"""Test that initialization fails without tokenizer."""
with pytest.raises(ValueError, match="model tokenizer must be passed"):
TestThinkingReasoningParser(None) # type: ignore[arg-type]
TestThinkingReasoningParser(None)
def test_initialization_with_missing_tokens(self, test_tokenizer):
"""Test that initialization fails when tokens are not in vocabulary."""
tests/renderers/test_cohere.py
+50 -50
View File
@@ -81,7 +81,7 @@ class TestRoleToMelody:
# reject non-strings — any exception type is acceptable as long
# as we don't silently produce a malformed prompt.
with pytest.raises((AttributeError, TypeError, ValueError)):
_role_to_melody(None) # type: ignore[arg-type]
_role_to_melody(None)
# ======================================================================
@@ -135,7 +135,7 @@ class TestNormalizeToolCall:
def test_invalid_type_rejected(self):
with pytest.raises(TypeError, match="Unexpected tool_call value"):
_normalize_tool_call(42) # type: ignore[arg-type]
_normalize_tool_call(42)
# ======================================================================
@@ -249,7 +249,7 @@ class TestDocumentToMelody:
def test_invalid_type_rejected(self):
with pytest.raises(TypeError, match="Unsupported document type"):
_document_to_melody(42) # type: ignore[arg-type]
_document_to_melody(42)
# ======================================================================
@@ -302,7 +302,7 @@ class TestToolToMelody:
def test_invalid_type_rejected(self):
with pytest.raises(TypeError, match="Unsupported tool type"):
_tool_to_melody(42) # type: ignore[arg-type]
_tool_to_melody(42)
# ======================================================================
@@ -418,7 +418,7 @@ class TestConversationToMelody:
{"role": "user", "content": "hi"},
{"role": "assistant", "content": "hello"},
]
out = _conversation_to_melody_messages(conv) # type: ignore[arg-type]
out = _conversation_to_melody_messages(conv)
assert out == [
{
"role": "user",
@@ -443,7 +443,7 @@ class TestConversationToMelody:
"reasoning": "thoughts",
}
]
out = _conversation_to_melody_messages(conv) # type: ignore[arg-type]
out = _conversation_to_melody_messages(conv)
assert out[0]["content"] == [
{"type": MelodyContentType.THINKING, "thinking": "thoughts"},
{"type": MelodyContentType.TEXT, "text": "answer"},
@@ -457,7 +457,7 @@ class TestConversationToMelody:
"reasoning_content": "thoughts",
}
]
out = _conversation_to_melody_messages(conv) # type: ignore[arg-type]
out = _conversation_to_melody_messages(conv)
assert out[0]["content"][0] == {
"type": MelodyContentType.THINKING,
"thinking": "thoughts",
@@ -474,7 +474,7 @@ class TestConversationToMelody:
"reasoning": "should be ignored",
}
]
out = _conversation_to_melody_messages(conv) # type: ignore[arg-type]
out = _conversation_to_melody_messages(conv)
assert out[0]["content"] == [{"type": MelodyContentType.TEXT, "text": "hi"}]
def test_tool_calls_normalized(self):
@@ -491,7 +491,7 @@ class TestConversationToMelody:
],
}
]
out = _conversation_to_melody_messages(conv) # type: ignore[arg-type]
out = _conversation_to_melody_messages(conv)
assert out[0]["tool_calls"] == [
{"id": "c1", "name": "f", "parameters": '{"a":1}'}
]
@@ -504,7 +504,7 @@ class TestConversationToMelody:
"tool_call_id": "c1",
}
]
out = _conversation_to_melody_messages(conv) # type: ignore[arg-type]
out = _conversation_to_melody_messages(conv)
assert out[0]["tool_call_id"] == "c1"
def test_messages_citations_attached_by_index(self):
@@ -531,20 +531,20 @@ class TestConversationToMelody:
}
]
}
out = _conversation_to_melody_messages(conv, citations) # type: ignore[arg-type]
out = _conversation_to_melody_messages(conv, citations)
assert "citations" not in out[0]
assert out[1]["citations"] == citations[1]
def test_messages_citations_none_is_a_no_op(self):
conv = [{"role": "assistant", "content": "a"}]
out = _conversation_to_melody_messages(conv, None) # type: ignore[arg-type]
out = _conversation_to_melody_messages(conv, None)
assert "citations" not in out[0]
def test_messages_citations_missing_index_is_a_no_op(self):
# A ``messages_citations`` dict whose key doesn't hit any
# message must not attach anything (and must not raise).
conv = [{"role": "assistant", "content": "a"}]
out = _conversation_to_melody_messages(conv, {5: [{"anything": 1}]}) # type: ignore[arg-type]
out = _conversation_to_melody_messages(conv, {5: [{"anything": 1}]})
assert "citations" not in out[0]
def test_tool_message_v2_content_replaces_default_content(self):
@@ -575,7 +575,7 @@ class TestConversationToMelody:
]
}
out = _conversation_to_melody_messages(
conv, # type: ignore[arg-type]
conv,
tool_message_v2_content=v2_content,
)
# User message (no forwarded entry) still goes through the
@@ -597,7 +597,7 @@ class TestConversationToMelody:
]
}
out = _conversation_to_melody_messages(
conv, # type: ignore[arg-type]
conv,
tool_message_v2_content=v2_content,
)
assert out[0]["content"] == [
@@ -607,7 +607,7 @@ class TestConversationToMelody:
def test_tool_message_v2_content_none_falls_back_to_default(self):
conv = [{"role": "tool", "tool_call_id": "c", "content": "plain text"}]
out = _conversation_to_melody_messages(
conv, # type: ignore[arg-type]
conv,
tool_message_v2_content=None,
)
assert out[0]["content"] == [{"type": "text", "text": "plain text"}]
@@ -626,7 +626,7 @@ class TestBuildRenderConfig:
# Bare kwargs -> cmd4 (the current Command A+ prompt format).
# Mirrors ``_DEFAULT_FORMAT`` in ``vllm/renderers/cohere.py`` and
# the ``--cohere-format`` CLI default.
fmt, cfg = _build_render_config(self._conv(), {}) # type: ignore[arg-type]
fmt, cfg = _build_render_config(self._conv(), {})
assert fmt == "cmd4"
assert cfg["use_jinja"] is True
assert isinstance(cfg["messages"], list)
@@ -634,12 +634,12 @@ class TestBuildRenderConfig:
assert "additional_template_fields" not in cfg
def test_explicit_cmd3(self):
fmt, cfg = _build_render_config(self._conv(), {"cohere_format": "cmd3"}) # type: ignore[arg-type]
fmt, cfg = _build_render_config(self._conv(), {"cohere_format": "cmd3"})
assert fmt == "cmd3"
def test_invalid_format_raises(self):
with pytest.raises(ValueError, match="Invalid cohere_format"):
_build_render_config(self._conv(), {"cohere_format": "cmd5"}) # type: ignore[arg-type]
_build_render_config(self._conv(), {"cohere_format": "cmd5"})
def test_documents_converted(self):
_, cfg = _build_render_config(
@@ -650,7 +650,7 @@ class TestBuildRenderConfig:
{"id": "d1", "data": {"text": "wrapped"}},
]
},
) # type: ignore[arg-type]
)
assert cfg["documents"] == [
{"text": "doc text"},
{"id": "d1", "text": "wrapped"},
@@ -665,7 +665,7 @@ class TestBuildRenderConfig:
{"type": "function", "function": {"name": "preferred"}}
],
},
) # type: ignore[arg-type]
)
names = [t["name"] for t in cfg["available_tools"]]
assert names == ["preferred"]
@@ -673,30 +673,30 @@ class TestBuildRenderConfig:
_, cfg = _build_render_config(
self._conv(),
{"tools": [{"type": "function", "function": {"name": "from_tools"}}]},
) # type: ignore[arg-type]
)
assert [t["name"] for t in cfg["available_tools"]] == ["from_tools"]
@pytest.mark.parametrize("value", ["enabled", "disabled"])
def test_reasoning_type_direct(self, value):
_, cfg = _build_render_config(self._conv(), {"reasoning_type": value}) # type: ignore[arg-type]
_, cfg = _build_render_config(self._conv(), {"reasoning_type": value})
assert cfg["reasoning_type"] == value
def test_thinking_dict_shorthand_resolves_reasoning_type(self):
_, cfg = _build_render_config(self._conv(), {"thinking": {"type": "enabled"}}) # type: ignore[arg-type]
_, cfg = _build_render_config(self._conv(), {"thinking": {"type": "enabled"}})
assert cfg["reasoning_type"] == "enabled"
def test_thinking_shorthand_ignores_unknown_type(self):
_, cfg = _build_render_config(self._conv(), {"thinking": {"type": "auto"}}) # type: ignore[arg-type]
_, cfg = _build_render_config(self._conv(), {"thinking": {"type": "auto"}})
assert "reasoning_type" not in cfg
def test_dev_instruction_forwarded(self):
_, cfg = _build_render_config(self._conv(), {"dev_instruction": "be brief"}) # type: ignore[arg-type]
_, cfg = _build_render_config(self._conv(), {"dev_instruction": "be brief"})
assert cfg["dev_instruction"] == "be brief"
def test_response_format_json_object_sets_json_mode(self):
_, cfg = _build_render_config(
self._conv(), {"response_format": {"type": "json_object"}}
) # type: ignore[arg-type]
)
assert cfg["json_mode"] is True
assert "json_schema" not in cfg
@@ -705,7 +705,7 @@ class TestBuildRenderConfig:
_, cfg = _build_render_config(
self._conv(),
{"response_format": {"type": "json_schema", "schema": schema}},
) # type: ignore[arg-type]
)
# JSON-encoded for melody (string-only schema field).
assert cfg["json_schema"] == json.dumps(schema)
@@ -721,35 +721,35 @@ class TestBuildRenderConfig:
"schema": {"schema": inner},
}
},
) # type: ignore[arg-type]
)
assert cfg["json_schema"] == json.dumps(inner)
def test_json_schema_kwarg_direct(self):
# Caller can also pass ``json_schema`` directly, both as dict and
# as a pre-stringified value.
_, cfg = _build_render_config(self._conv(), {"json_schema": {"a": 1}}) # type: ignore[arg-type]
_, cfg = _build_render_config(self._conv(), {"json_schema": {"a": 1}})
assert cfg["json_schema"] == '{"a": 1}'
_, cfg = _build_render_config(
self._conv(), {"json_schema": "raw-string-schema"}
) # type: ignore[arg-type]
)
assert cfg["json_schema"] == "raw-string-schema"
def test_json_mode_kwarg_overrides(self):
_, cfg = _build_render_config(self._conv(), {"json_mode": True}) # type: ignore[arg-type]
_, cfg = _build_render_config(self._conv(), {"json_mode": True})
assert cfg["json_mode"] is True
def test_cmd3_safety_mode_lowercased(self):
_, cfg = _build_render_config(
self._conv(),
{"cohere_format": "cmd3", "safety_mode": "CONTEXTUAL"},
) # type: ignore[arg-type]
)
assert cfg["safety_mode"] == "contextual"
def test_cmd3_citation_quality_direct(self):
_, cfg = _build_render_config(
self._conv(),
{"cohere_format": "cmd3", "citation_quality": "ACCURATE"},
) # type: ignore[arg-type]
)
assert cfg["citation_quality"] == "accurate"
def test_cmd3_citation_quality_derived_from_citation_options(self):
@@ -758,20 +758,20 @@ class TestBuildRenderConfig:
_, cfg = _build_render_config(
self._conv(),
{"cohere_format": "cmd3", "citation_options": {"mode": "accurate"}},
) # type: ignore[arg-type]
)
assert cfg["citation_quality"] == "on"
_, cfg = _build_render_config(
self._conv(),
{"cohere_format": "cmd3", "citation_options": {"mode": "off"}},
) # type: ignore[arg-type]
)
assert cfg["citation_quality"] == "off"
def test_cmd3_skip_preamble_forwarded(self):
_, cfg = _build_render_config(
self._conv(),
{"cohere_format": "cmd3", "skip_preamble": True},
) # type: ignore[arg-type]
)
assert cfg["skip_preamble"] is True
def test_cmd3_no_grounding_field(self):
@@ -779,7 +779,7 @@ class TestBuildRenderConfig:
_, cfg = _build_render_config(
self._conv(),
{"cohere_format": "cmd3", "grounding": "fast"},
) # type: ignore[arg-type]
)
assert "grounding" not in cfg
@pytest.mark.parametrize(
@@ -800,7 +800,7 @@ class TestBuildRenderConfig:
_, cfg = _build_render_config(
self._conv(),
{"cohere_format": "cmd4", "grounding": raw},
) # type: ignore[arg-type]
)
assert cfg["grounding"] == expected
@pytest.mark.parametrize(
@@ -818,7 +818,7 @@ class TestBuildRenderConfig:
"cohere_format": "cmd4",
"citation_options": {"mode": mode},
},
) # type: ignore[arg-type]
)
assert cfg["grounding"] == expected
def test_cmd4_grounding_rejects_unknown_value(self):
@@ -826,7 +826,7 @@ class TestBuildRenderConfig:
_build_render_config(
self._conv(),
{"cohere_format": "cmd4", "grounding": "foobar"},
) # type: ignore[arg-type]
)
def test_cmd4_platform_instruction(self):
_, cfg = _build_render_config(
@@ -835,7 +835,7 @@ class TestBuildRenderConfig:
"cohere_format": "cmd4",
"platform_instruction": "do this",
},
) # type: ignore[arg-type]
)
assert cfg["platform_instruction"] == "do this"
def test_cmd4_no_safety_mode_field(self):
@@ -847,7 +847,7 @@ class TestBuildRenderConfig:
"safety_mode": "contextual",
"citation_quality": "on",
},
) # type: ignore[arg-type]
)
assert "safety_mode" not in cfg
assert "citation_quality" not in cfg
@@ -862,7 +862,7 @@ class TestBuildRenderConfig:
"my_var": "x",
"documents": ["doc"], # consumed, must NOT leak through
},
) # type: ignore[arg-type]
)
extras = cfg["additional_template_fields"]
assert extras == {"reasoning_effort": "low", "my_var": "x"}
# Sanity: the consumed key still produced its dedicated config slot.
@@ -875,7 +875,7 @@ class TestBuildRenderConfig:
_, cfg = _build_render_config(
self._conv(),
{"template_id": "tpl1"},
) # type: ignore[arg-type]
)
assert cfg["template_id"] == "tpl1"
# use_jinja is always True, regardless of caller input.
assert cfg["use_jinja"] is True
@@ -896,7 +896,7 @@ class TestBuildRenderConfig:
_build_render_config(
self._conv(),
{cohere_only_key: "raw {{ jinja }}"},
) # type: ignore[arg-type]
)
@pytest.mark.parametrize("cohere_only_key", ["template_jinja", "template"])
def test_cohere_only_template_kwargs_none_is_tolerated(self, cohere_only_key):
@@ -906,7 +906,7 @@ class TestBuildRenderConfig:
_, cfg = _build_render_config(
self._conv(),
{cohere_only_key: None},
) # type: ignore[arg-type]
)
assert cohere_only_key not in cfg
assert "additional_template_fields" not in cfg
@@ -918,7 +918,7 @@ class TestBuildRenderConfig:
self._conv(),
{},
"raw {{ jinja }}",
) # type: ignore[arg-type]
)
assert cfg["template_jinja"] == "raw {{ jinja }}"
assert cfg["use_jinja"] is True
@@ -927,7 +927,7 @@ class TestBuildRenderConfig:
self._conv(),
{},
None,
) # type: ignore[arg-type]
)
assert "template_jinja" not in cfg
def test_tool_message_v2_content_key_picked_up_from_kwargs(self):
@@ -954,7 +954,7 @@ class TestBuildRenderConfig:
]
}
}
_, cfg = _build_render_config(conv, kwargs) # type: ignore[arg-type]
_, cfg = _build_render_config(conv, kwargs)
tool_msg = cfg["messages"][1]
assert tool_msg["content"] == [{"type": "document", "document": {"r": "k"}}]
# And the key must be *consumed* -- not surfaced as a stray
tests/renderers/test_completions.py
+1 -1
View File
@@ -139,7 +139,7 @@ class TestValidatePrompt:
with pytest.raises(TypeError, match="should be a list of integers"):
renderer.render_prompts(
_preprocess_prompt(renderer.model_config, [[1, 2], ["foo", "bar"]]) # type: ignore[arg-type]
_preprocess_prompt(renderer.model_config, [[1, 2], ["foo", "bar"]])
)
tests/samplers/test_beam_search.py
+3 -3
View File
@@ -113,7 +113,7 @@ def test_beam_search_abort_returns_partial_outputs_and_continues_other_prompts(
{"type": "token", "prompt_token_ids": [token]} for token in [prompt_token, 2]
]
params = BeamSearchParams(beam_width=2, max_tokens=3)
outputs = llm.beam_search(prompts, params) # type: ignore[arg-type]
outputs = llm.beam_search(prompts, params)
aborted, normal = outputs
expected_tokens = (
@@ -326,7 +326,7 @@ def test_beam_search_structured_output(
llm = LLM(
model=model,
dtype=dtype, # type: ignore[arg-type]
dtype=dtype,
max_model_len=512,
structured_outputs_config=dict(
backend="xgrammar",
@@ -345,7 +345,7 @@ def test_beam_search_structured_output(
"Generate a JSON object for a person with name and age:",
]
outputs = llm.beam_search(prompts, params) # type: ignore[arg-type]
outputs = llm.beam_search(prompts, params)
assert len(outputs) == len(prompts)
for output in outputs:
tests/samplers/test_beam_search_online.py
+2 -2
View File
@@ -57,8 +57,8 @@ class _EngineClient:
class _Serving(BeamSearchOnlineMixin):
renderer = _Renderer() # type: ignore[assignment]
engine_client = _EngineClient() # type: ignore[assignment]
renderer = _Renderer()
engine_client = _EngineClient()
@pytest.mark.asyncio
tests/standalone_tests/lazy_imports.py
+1 -1
View File
@@ -17,6 +17,6 @@ module_names = ["torch._inductor.async_compile", "cv2"]
# if we import any modules during `import vllm`, there would be a
# hard error and nice stacktrace on the first import.
for module_name in module_names:
sys.modules[module_name] = None # type: ignore[assignment]
sys.modules[module_name] = None
import vllm # noqa
tests/test_audio_media_size_precheck.py
+1 -1
View File
@@ -144,6 +144,6 @@ async def test_audio_connector_threads_byte_limit_to_http_reader(
assert kwargs["max_bytes"] == MiB_bytes
raise VLLMValidationError("Maximum file size exceeded")
connector = MediaConnector(connection=_Connection()) # type: ignore[arg-type]
connector = MediaConnector(connection=_Connection())
with pytest.raises(VLLMValidationError, match="Maximum file size exceeded"):
await connector.fetch_audio_async("https://example.test/audio")
tests/tokenizers_/test_mistral.py
+5 -5
View File
@@ -975,7 +975,7 @@ class TestMistralTokenizer:
mistral_tokenizer: MistralTokenizer,
):
assert mistral_tokenizer.batch_decode(
[[]], # type: ignore[arg-type]
[[]],
) == [""]
def test_convert_tokens_to_string(self, mistral_tokenizer: MistralTokenizer):
@@ -2246,7 +2246,7 @@ class TestMistralTokenizer:
]
output = mistral_tokenizer.apply_chat_template(
messages, # type: ignore[arg-type]
messages,
add_generation_prompt=True,
)
decoded = mistral_tokenizer.tokenizer.decode(output, SpecialTokenPolicy.KEEP)
@@ -2352,7 +2352,7 @@ def test_v15_apply_chat_template_passes_reasoning_effort_high(
)
v15_mistral_tokenizer.apply_chat_template(
messages=_PASSTHROUGH_MESSAGES, # type: ignore[arg-type]
messages=_PASSTHROUGH_MESSAGES,
reasoning_effort="high",
)
@@ -2375,7 +2375,7 @@ def test_v15_apply_chat_template_passes_reasoning_effort_none_by_default(
fake_apply_chat_template,
)
v15_mistral_tokenizer.apply_chat_template(messages=_PASSTHROUGH_MESSAGES) # type: ignore[arg-type]
v15_mistral_tokenizer.apply_chat_template(messages=_PASSTHROUGH_MESSAGES)
assert "reasoning_effort" in captured_kwargs[-1]
assert captured_kwargs[-1]["reasoning_effort"] is None
@@ -2398,7 +2398,7 @@ def test_pre_v15_apply_chat_template_omits_reasoning_effort(
)
v13_mistral_tokenizer.apply_chat_template(
messages=_PASSTHROUGH_MESSAGES, # type: ignore[arg-type]
messages=_PASSTHROUGH_MESSAGES,
reasoning_effort="high",
)
tests/tokenizers_/test_registry.py
+1 -1
View File
@@ -205,7 +205,7 @@ def test_cached_tokenizer_from_config_registers_local_config(tmp_path: Path):
side_effect=lambda tokenizer: tokenizer,
),
):
tokenizer = cached_tokenizer_from_config(model_config) # type: ignore[arg-type]
tokenizer = cached_tokenizer_from_config(model_config)
assert tokenizer.is_fast is True
finally:
tests/tool_parsers/test_ernie45_moe_tool_parser.py
+2 -2
View File
@@ -54,7 +54,7 @@ def test_extract_tool_calls_no_tools(ernie45_tool_parser):
model_output = "This is a test"
extracted_tool_calls = ernie45_tool_parser.extract_tool_calls(
model_output, request=None
) # type: ignore[arg-type]
)
assert not extracted_tool_calls.tools_called
assert extracted_tool_calls.tool_calls == []
assert extracted_tool_calls.content == model_output
@@ -163,7 +163,7 @@ def test_extract_tool_calls(
):
extracted_tool_calls = ernie45_tool_parser.extract_tool_calls(
model_output, request=None
) # type: ignore[arg-type]
)
assert extracted_tool_calls.tools_called
assert_tool_calls(extracted_tool_calls.tool_calls, expected_tool_calls)
tests/tool_parsers/test_jamba_tool_parser.py
+3 -3
View File
@@ -78,7 +78,7 @@ def stream_delta_message_generator(
previous_token_ids,
current_token_ids,
delta_token_ids,
request=None, # type: ignore[arg-type]
request=None,
)
if delta_message:
yield delta_message
@@ -95,7 +95,7 @@ def test_extract_tool_calls_no_tools(jamba_tool_parser):
model_output = "This is a test"
extracted_tool_calls = jamba_tool_parser.extract_tool_calls(
model_output, request=None
) # type: ignore[arg-type]
)
assert not extracted_tool_calls.tools_called
assert extracted_tool_calls.tool_calls == []
assert extracted_tool_calls.content == model_output
@@ -166,7 +166,7 @@ def test_extract_tool_calls(
):
extracted_tool_calls = jamba_tool_parser.extract_tool_calls(
model_output, request=None
) # type: ignore[arg-type]
)
assert extracted_tool_calls.tools_called
assert_tool_calls(extracted_tool_calls.tool_calls, expected_tool_calls)
tests/tool_parsers/test_qwen3coder_tool_parser.py
+1 -1
View File
@@ -225,7 +225,7 @@ def test_extract_tool_calls_no_tools(qwen3_tool_parser):
model_output = "This is a test response without any tool calls"
extracted_tool_calls = qwen3_tool_parser.extract_tool_calls(
model_output, request=None
) # type: ignore[arg-type]
)
assert not extracted_tool_calls.tools_called
assert extracted_tool_calls.tool_calls == []
assert extracted_tool_calls.content == model_output
tests/tool_parsers/test_step3p5_tool_parser.py
+1 -1
View File
@@ -170,7 +170,7 @@ def test_extract_tool_calls_no_tools(step3p5_tool_parser):
model_output = "This is a test response without any tool calls"
extracted_tool_calls = step3p5_tool_parser.extract_tool_calls(
model_output, request=None
) # type: ignore[arg-type]
)
assert not extracted_tool_calls.tools_called
assert extracted_tool_calls.tool_calls == []
assert extracted_tool_calls.content == model_output
tests/tool_parsers/test_xlam_tool_parser.py
+4 -4
View File
@@ -100,7 +100,7 @@ def test_extract_tool_calls_no_tools(xlam_tool_parser):
model_output = "This is a test"
extracted_tool_calls = xlam_tool_parser.extract_tool_calls(
model_output, request=None
) # type: ignore[arg-type]
)
assert not extracted_tool_calls.tools_called
assert extracted_tool_calls.tool_calls == []
assert extracted_tool_calls.content == model_output
@@ -225,7 +225,7 @@ def test_extract_tool_calls(
):
extracted_tool_calls = xlam_tool_parser.extract_tool_calls(
model_output, request=None
) # type: ignore[arg-type]
)
assert extracted_tool_calls.tools_called
assert_tool_calls(extracted_tool_calls.tool_calls, expected_tool_calls)
@@ -263,7 +263,7 @@ def test_extract_tool_calls_list_structure(
"""Test extraction of tool calls when the model outputs a list-structured tool call.""" # noqa: E501
extracted_tool_calls = xlam_tool_parser.extract_tool_calls(
model_output, request=None
) # type: ignore[arg-type]
)
assert extracted_tool_calls.tools_called
assert_tool_calls(extracted_tool_calls.tool_calls, expected_tool_calls)
@@ -550,7 +550,7 @@ def test_extract_tool_calls_non_ascii(xlam_tool_parser, xlam_tokenizer, streamin
if delta.tool_calls and delta.tool_calls[0].function.arguments
)
else:
extracted = xlam_tool_parser.extract_tool_calls(model_output, request=None) # type: ignore[arg-type]
extracted = xlam_tool_parser.extract_tool_calls(model_output, request=None)
args = "".join(tc.function.arguments for tc in extracted.tool_calls)
assert "北京" in args
tests/tool_use/test_gemma4_responses_adjust_request.py
+1 -1
View File
@@ -148,7 +148,7 @@ def test_tool_parser_adjust_request_builds_valid_response_text_config() -> None:
description string that the previous two-step construction injected.
"""
parser = ToolParser.__new__(ToolParser)
parser.model_tokenizer = None # type: ignore[assignment]
parser.model_tokenizer = None
request = _build_responses_request(tool_choice="required")
ToolParser.adjust_request(parser, request)
tests/tool_use/test_kimi_k3_tool_parser.py
+1 -1
View File
@@ -370,7 +370,7 @@ def test_streaming_split_markers_do_not_leak():
assert len(tool_deltas) == 1
assert tool_deltas[0].function is not None
assert tool_deltas[0].function.name == "calc"
assert json.loads(tool_deltas[0].function.arguments) == {"x": 1} # type: ignore[arg-type]
assert json.loads(tool_deltas[0].function.arguments) == {"x": 1}
def test_tool_call_ids_are_unique_across_messages():
tests/transformers_utils/processors/test_voxtral.py
+1 -1
View File
@@ -98,7 +98,7 @@ def test_fetch_audio_rejects_unsupported_type(
feature_extractor: MistralCommonFeatureExtractor,
):
with pytest.raises(TypeError, match="only a numpy array"):
feature_extractor.fetch_audio(42) # type: ignore[arg-type]
feature_extractor.fetch_audio(42)
def test_fetch_audio_str_delegates_to_load_audio(
tests/transformers_utils/test_processor.py
+2 -2
View File
@@ -44,7 +44,7 @@ class _ProcWithUnpack:
def test_get_processor_kwargs_from_processor_unpack_path_returns_full_union():
proc = _ProcWithUnpack()
keys = get_processor_kwargs_keys(get_processor_kwargs_type(proc)) # type: ignore[arg-type]
keys = get_processor_kwargs_keys(get_processor_kwargs_type(proc))
_assert_has_all_expected(keys)
@@ -63,5 +63,5 @@ def test_get_processor_kwargs_from_processor_module_scan_returns_full_union():
assert hasattr(mod, "_FakeProcessorKwargs")
proc = _ProcWithoutUnpack()
keys = get_processor_kwargs_keys(get_processor_kwargs_type(proc)) # type: ignore[arg-type]
keys = get_processor_kwargs_keys(get_processor_kwargs_type(proc))
_assert_has_all_expected(keys)
tests/utils.py
+1 -1
View File
@@ -830,7 +830,7 @@ class RemoteOpenAIServerCustom(RemoteOpenAIServer):
self.proc: Process = cast(Any, ctx).Process(
target=_run_in_new_process_group,
args=(self.child_process_fxn, env_dict, model, vllm_serve_args),
) # type: ignore[assignment]
)
with _temporarily_sanitized_pythonpath_env():
self.proc.start()
tests/utils_/test_argparse_utils.py
+9 -9
View File
@@ -476,32 +476,32 @@ def test_compilation_config_mode_validator():
"""Test that CompilationConfig.mode field validator converts strings to integers."""
from vllm.config.compilation import CompilationConfig, CompilationMode
config = CompilationConfig(mode=0) # type: ignore[arg-type]
config = CompilationConfig(mode=0)
assert config.mode == CompilationMode.NONE
config = CompilationConfig(mode=3) # type: ignore[arg-type]
config = CompilationConfig(mode=3)
assert config.mode == CompilationMode.VLLM_COMPILE
config = CompilationConfig(mode="NONE") # type: ignore[arg-type]
config = CompilationConfig(mode="NONE")
assert config.mode == CompilationMode.NONE
config = CompilationConfig(mode="STOCK_TORCH_COMPILE") # type: ignore[arg-type]
config = CompilationConfig(mode="STOCK_TORCH_COMPILE")
assert config.mode == CompilationMode.STOCK_TORCH_COMPILE
config = CompilationConfig(mode="DYNAMO_TRACE_ONCE") # type: ignore[arg-type]
config = CompilationConfig(mode="DYNAMO_TRACE_ONCE")
assert config.mode == CompilationMode.DYNAMO_TRACE_ONCE
config = CompilationConfig(mode="VLLM_COMPILE") # type: ignore[arg-type]
config = CompilationConfig(mode="VLLM_COMPILE")
assert config.mode == CompilationMode.VLLM_COMPILE
config = CompilationConfig(mode="none") # type: ignore[arg-type]
config = CompilationConfig(mode="none")
assert config.mode == CompilationMode.NONE
config = CompilationConfig(mode="vllm_compile") # type: ignore[arg-type]
config = CompilationConfig(mode="vllm_compile")
assert config.mode == CompilationMode.VLLM_COMPILE
with pytest.raises(ValidationError, match="Invalid compilation mode"):
CompilationConfig(mode="INVALID_MODE") # type: ignore[arg-type]
CompilationConfig(mode="INVALID_MODE")
def test_flat_product():
tests/v1/attention/test_mla_prefill_selector.py
+2 -2
View File
@@ -511,14 +511,14 @@ class TestMLAPrefillBackendParsing:
def test_valid_string_parses_to_enum(self):
config = AttentionConfig(
mla_prefill_backend="FLASH_ATTN", # type: ignore[arg-type]
mla_prefill_backend="FLASH_ATTN",
)
assert config.mla_prefill_backend == MLAPrefillBackendEnum.FLASH_ATTN
def test_invalid_string_raises_error(self):
with pytest.raises(ValueError, match="Unknown MLA prefill backend"):
AttentionConfig(
mla_prefill_backend="NONEXISTENT", # type: ignore[arg-type]
mla_prefill_backend="NONEXISTENT",
)
tests/v1/attention/test_sparse_mla_backends.py
+3 -3
View File
@@ -3575,7 +3575,7 @@ def test_flashmla_fp8_paths_accept_decode_subset(monkeypatch, use_mixed_batch: b
if use_mixed_batch:
fp8_metadata = FlashMLASparseMetadata.FP8KernelMetadata(
scheduler_metadata=object(), # type: ignore[arg-type]
scheduler_metadata=object(),
dummy_block_table=torch.empty(1, 1, dtype=torch.int32, device=DEVICE_TYPE),
cache_lens=torch.empty(1, dtype=torch.int32, device=DEVICE_TYPE),
)
@@ -3588,7 +3588,7 @@ def test_flashmla_fp8_paths_accept_decode_subset(monkeypatch, use_mixed_batch: b
num_prefill_tokens=num_batch_tokens - num_decode_tokens,
decode=FP8Meta.Decode(
seq_lens=torch.empty(1, dtype=torch.int32, device=DEVICE_TYPE),
kernel_metadata=object(), # type: ignore[arg-type]
kernel_metadata=object(),
decode_query_len=num_decode_tokens,
),
prefill=FP8Meta.Prefill(
@@ -3774,7 +3774,7 @@ def test_fp8_mixed_batch_dcp_neutralizes_empty_rows(monkeypatch):
metadata = SimpleNamespace(
fp8_extra_metadata=FlashMLASparseMetadata.FP8KernelMetadata(
scheduler_metadata=object(), # type: ignore[arg-type]
scheduler_metadata=object(),
dummy_block_table=torch.empty(1, 1, dtype=torch.int32, device=DEVICE_TYPE),
cache_lens=torch.empty(1, dtype=torch.int32, device=DEVICE_TYPE),
),
tests/v1/cudagraph/test_cudagraph_manager.py
+2 -2
View File
@@ -134,7 +134,7 @@ def test_piecewise_capture_uses_pcp_dummy_slot_mappings():
device=torch.device("cpu"),
max_num_reqs=num_reqs,
max_num_tokens=num_tokens,
block_tables=pcp_block_tables, # type: ignore[arg-type]
block_tables=pcp_block_tables,
)
block_tables = MagicMock()
@@ -224,7 +224,7 @@ def test_speculator_capture_preserves_decode_query_bounds(
)
manager = SpeculatorCudaGraphManager.__new__(SpeculatorCudaGraphManager)
manager.max_num_reqs = num_reqs
manager.vllm_config = SimpleNamespace( # type: ignore[assignment]
manager.vllm_config = SimpleNamespace(
parallel_config=SimpleNamespace(data_parallel_size=1)
)
buffers = InputBuffers(num_reqs, num_tokens, torch.device("cpu"))

Some files were not shown because too many files have changed in this diff Show More