[MyPy][2/N] Fix mypy errors in small tests/ dirs (#51043)
Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
co-authored by
Harry Mellor
Claude Opus 5.5
parent
a96ee5935d
commit
5dd693122c
@@ -17,6 +17,7 @@ import vllm.config as vllm_config_module
|
||||
from vllm.config import VllmConfig, set_current_vllm_config
|
||||
from vllm.model_executor.layers.fused_moe import utils as fused_moe_utils
|
||||
from vllm.model_executor.layers.fused_moe.layer import determine_expert_counts
|
||||
from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
|
||||
from vllm.model_executor.layers.quantization.fp8 import Fp8Config
|
||||
from vllm.model_executor.layers.quantization.quark.quark import QuarkConfig
|
||||
from vllm.model_executor.layers.quantization.utils.config_utils import (
|
||||
@@ -140,10 +141,15 @@ def get_deepseek_v4_quark_config(exclude: list[str]) -> dict[str, Any]:
|
||||
return quantization_config
|
||||
|
||||
|
||||
def _stub_quant_config() -> QuantizationConfig:
|
||||
return cast(QuantizationConfig, object())
|
||||
|
||||
|
||||
def get_fse_test_model_config(
|
||||
model_type: str,
|
||||
quantization_config: dict[str, Any],
|
||||
) -> tuple[object, type[nn.Module]]:
|
||||
config: object
|
||||
if model_type == "minimax_m3":
|
||||
config = MiniMaxM3TextConfig(
|
||||
hidden_size=128,
|
||||
@@ -301,7 +307,7 @@ def test_resolve_layer_fused_shared_expert_skips_compatibility_when_disabled(
|
||||
)
|
||||
|
||||
assert not fused_moe_utils.resolve_layer_fused_shared_expert(
|
||||
object(), "model.layers.0.mlp"
|
||||
_stub_quant_config(), "model.layers.0.mlp"
|
||||
)
|
||||
|
||||
|
||||
@@ -316,7 +322,7 @@ def test_resolve_layer_fused_shared_expert_normalizes_unavailable_aiter(
|
||||
|
||||
assert (
|
||||
fused_moe_utils.resolve_layer_fused_shared_expert(
|
||||
object(), "model.layers.0.mlp"
|
||||
_stub_quant_config(), "model.layers.0.mlp"
|
||||
)
|
||||
is False
|
||||
)
|
||||
@@ -325,7 +331,7 @@ def test_resolve_layer_fused_shared_expert_normalizes_unavailable_aiter(
|
||||
def test_resolve_layer_fused_shared_expert_passes_module_prefixes(
|
||||
monkeypatch: pytest.MonkeyPatch,
|
||||
) -> None:
|
||||
quant_config = object()
|
||||
quant_config = _stub_quant_config()
|
||||
monkeypatch.setattr(
|
||||
fused_moe_utils.rocm_aiter_ops,
|
||||
"is_fusion_moe_shared_experts_enabled",
|
||||
@@ -374,7 +380,7 @@ def test_resolve_layer_fused_shared_expert_rejects_incompatible_quantization(
|
||||
)
|
||||
|
||||
assert not fused_moe_utils.resolve_layer_fused_shared_expert(
|
||||
object(), "model.layers.0.mlp"
|
||||
_stub_quant_config(), "model.layers.0.mlp"
|
||||
)
|
||||
assert "shared experts are excluded" in caplog.text
|
||||
|
||||
@@ -407,7 +413,7 @@ def test_deepseek_v4_shared_expert_fse_uses_mtp_quantization_config_prefix(
|
||||
)
|
||||
|
||||
compatible, reason = is_shared_expert_quant_fse_compatible(
|
||||
DeepseekV4Config(),
|
||||
DeepseekV4Config(), # type: ignore[arg-type]
|
||||
"model.layers.2.ffn.experts",
|
||||
"model.layers.2.ffn.shared_experts",
|
||||
)
|
||||
@@ -818,7 +824,7 @@ def test_models_fse_init(
|
||||
model_type, quantization_config
|
||||
)
|
||||
vllm_config = VllmConfig()
|
||||
vllm_config.model_config = SimpleNamespace(
|
||||
vllm_config.model_config = SimpleNamespace( # type: ignore[assignment]
|
||||
hf_config=config,
|
||||
hf_text_config=config,
|
||||
dtype=torch.bfloat16,
|
||||
@@ -887,7 +893,7 @@ def test_models_fse_init(
|
||||
DeepSeekV4MTP,
|
||||
)
|
||||
|
||||
vllm_config.speculative_config = SimpleNamespace(
|
||||
vllm_config.speculative_config = SimpleNamespace( # type: ignore[assignment]
|
||||
draft_model_config=SimpleNamespace(hf_config=config),
|
||||
method="mtp",
|
||||
parallel_drafting=False,
|
||||
@@ -1214,7 +1220,7 @@ def test_quark_packed_layer_config_must_match_global_config() -> None:
|
||||
|
||||
def test_non_quark_shared_expert_fse_is_incompatible() -> None:
|
||||
compatible, reason = is_shared_expert_quant_fse_compatible(
|
||||
object(),
|
||||
_stub_quant_config(),
|
||||
"model.layers.0.mlp.experts",
|
||||
"model.layers.0.mlp.shared_experts",
|
||||
)
|
||||
|
||||
@@ -46,7 +46,7 @@ def test_sparse_attention_refreshes_batch_state_inside_eager_segment(
|
||||
|
||||
with pytest.raises(BatchStateRefreshed):
|
||||
DeepseekV32Attention._sparse_indexer_and_attn(
|
||||
layer,
|
||||
layer, # type: ignore[arg-type]
|
||||
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,
|
||||
prefill=SimpleNamespace(chunks=[]) if num_prefills else None, # type: ignore[arg-type]
|
||||
)
|
||||
|
||||
|
||||
@@ -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()
|
||||
indexer_metadata.decode = object() # type: ignore[assignment]
|
||||
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,
|
||||
layer, # type: ignore[arg-type]
|
||||
torch.arange(2),
|
||||
torch.empty(2, 2),
|
||||
q_nope,
|
||||
|
||||
@@ -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)
|
||||
result = resolve_classifier_act_fn(model_config, act_fn=None) # type: ignore[arg-type]
|
||||
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)
|
||||
result = resolve_classifier_act_fn(None, act_fn=custom) # type: ignore[arg-type]
|
||||
assert result is custom
|
||||
|
||||
@@ -15,6 +15,7 @@ from vllm.model_executor.layers.pooler.tokwise.heads import (
|
||||
TokenEmbeddingPoolerHead,
|
||||
)
|
||||
from vllm.pooling_params import PoolingParams
|
||||
from vllm.tasks import PoolingTask
|
||||
from vllm.v1.pool.metadata import PoolingMetadata, PoolingStates
|
||||
|
||||
_HIDDEN = 16
|
||||
@@ -24,7 +25,7 @@ _BATCH = 3
|
||||
def _make_params(
|
||||
n: int,
|
||||
*,
|
||||
task: str = "embed",
|
||||
task: PoolingTask = "embed",
|
||||
dimensions: int | None = None,
|
||||
use_activation: bool | None = None,
|
||||
) -> list[PoolingParams]:
|
||||
@@ -320,6 +321,7 @@ class TestTokenEmbeddingPoolerHead:
|
||||
x = torch.randn(5, _HIDDEN)
|
||||
param = PoolingParams(task="token_embed")
|
||||
out = head.forward_chunk(x, param)
|
||||
assert out is not None
|
||||
assert out.dtype == torch.float16
|
||||
|
||||
def test_projector(self):
|
||||
@@ -328,6 +330,7 @@ class TestTokenEmbeddingPoolerHead:
|
||||
x = torch.randn(5, _HIDDEN)
|
||||
param = PoolingParams(task="token_embed")
|
||||
out = head.forward_chunk(x, param)
|
||||
assert out is not None
|
||||
assert out.shape == (5, 8)
|
||||
assert torch.allclose(out, proj(x))
|
||||
|
||||
@@ -336,6 +339,7 @@ class TestTokenEmbeddingPoolerHead:
|
||||
x = torch.randn(5, _HIDDEN)
|
||||
param = PoolingParams(task="token_embed", dimensions=4)
|
||||
out = head.forward_chunk(x, param)
|
||||
assert out is not None
|
||||
assert out.shape == (5, 4)
|
||||
assert torch.equal(out, x[..., :4])
|
||||
|
||||
@@ -360,6 +364,7 @@ class TestTokenEmbeddingPoolerHead:
|
||||
x = torch.randn(5, _HIDDEN)
|
||||
param = PoolingParams(task="token_embed", dimensions=4)
|
||||
out = head.forward_chunk(x, param)
|
||||
assert out is not None
|
||||
assert out.shape == (5, 4)
|
||||
assert torch.equal(out, proj(x)[..., :4])
|
||||
|
||||
@@ -368,6 +373,7 @@ class TestTokenEmbeddingPoolerHead:
|
||||
x = torch.randn(5, _HIDDEN)
|
||||
param = PoolingParams(task="token_embed", dimensions=4, use_activation=True)
|
||||
out = head.forward_chunk(x, param)
|
||||
assert out is not None
|
||||
assert out.shape == (5, 4)
|
||||
norms = torch.linalg.norm(out, dim=-1)
|
||||
assert torch.allclose(norms, torch.ones(5), atol=1e-5)
|
||||
@@ -416,6 +422,7 @@ class TestTokenClassifierPoolerHead:
|
||||
x = torch.randn(5, _HIDDEN)
|
||||
param = PoolingParams(task="token_classify")
|
||||
out = head.forward_chunk(x, param)
|
||||
assert out is not None
|
||||
assert out.dtype == torch.float16
|
||||
|
||||
def test_classifier(self):
|
||||
@@ -424,6 +431,7 @@ class TestTokenClassifierPoolerHead:
|
||||
x = torch.randn(5, _HIDDEN)
|
||||
param = PoolingParams(task="token_classify")
|
||||
out = head.forward_chunk(x, param)
|
||||
assert out is not None
|
||||
assert out.shape == (5, 3)
|
||||
assert torch.allclose(out, clf(x))
|
||||
|
||||
|
||||
@@ -26,6 +26,7 @@ from vllm.model_executor.layers.pooler.tokwise.methods import (
|
||||
get_tok_pooling_method,
|
||||
)
|
||||
from vllm.pooling_params import PoolingParams
|
||||
from vllm.tasks import PoolingTask
|
||||
from vllm.v1.pool.metadata import PoolingCursor, PoolingMetadata, PoolingStates
|
||||
|
||||
_CPU = torch.device("cpu")
|
||||
@@ -66,7 +67,7 @@ def _make_pooling_cursor(
|
||||
def _make_metadata(
|
||||
prompt_lens: list[int],
|
||||
*,
|
||||
tasks: list[str] | None = None,
|
||||
tasks: list[PoolingTask] | None = None,
|
||||
token_ids: list[list[int]] | None = None,
|
||||
pooling_params: list[PoolingParams] | None = None,
|
||||
num_scheduled_tokens: list[int] | None = None,
|
||||
@@ -319,7 +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):
|
||||
with set_current_vllm_config(config): # type: ignore[arg-type]
|
||||
pooler = DispatchPooler.for_seq_cls(pooler_config, classifier=classifier)
|
||||
|
||||
token_ids = [[2, 99, 3, 99], [4, 99], [5, 6]]
|
||||
|
||||
Reference in New Issue
Block a user