[MyPy][1/N] Fix mypy errors in some tests/ directories and enforce follow-imports=silent (#49570)

Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Martin Hickey
2026-07-30 05:02:51 -07:00
committed by GitHub
co-authored by Harry Mellor
parent 89d97d9a16
commit 38a267cdd5
30 changed files with 166 additions and 101 deletions
@@ -35,10 +35,12 @@ def _apply_qk_norm_rope(
q_by_head = q.view(*q.shape[:-1], q.shape[-1] // head_dim, head_dim)
q_by_head = q_norm.forward_native(q_by_head)
assert isinstance(q_by_head, torch.Tensor)
q = q_by_head.view(q.shape)
k_by_head = k.view(*k.shape[:-1], k.shape[-1] // head_dim, head_dim)
k_by_head = k_norm.forward_native(k_by_head)
assert isinstance(k_by_head, torch.Tensor)
k = k_by_head.view(k.shape)
q, k = rope.forward_native(positions, q, k)
+3 -4
View File
@@ -34,10 +34,9 @@ NUM_HEADS = [16]
@pytest.fixture
def _fp8_attention():
"""Create FP8-enabled MMEncoderAttention via config."""
from types import SimpleNamespace
from unittest.mock import patch
from unittest.mock import MagicMock, patch
from vllm.config import VllmConfig, set_current_vllm_config
from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config
from vllm.config.multimodal import MultiModalConfig
if not is_flashinfer_cudnn_fp8_prefill_attn_supported():
@@ -45,7 +44,7 @@ def _fp8_attention():
mm_config = MultiModalConfig(mm_encoder_attn_dtype="fp8")
vllm_config = VllmConfig()
vllm_config.model_config = SimpleNamespace(multimodal_config=mm_config)
vllm_config.model_config = MagicMock(spec=ModelConfig, multimodal_config=mm_config)
# MMEncoderAttention reads torch.get_default_dtype() during init
# to determine the output dtype. In real model loading this is bf16.
+5 -6
View File
@@ -4,8 +4,7 @@
import contextlib
import json
from types import SimpleNamespace
from unittest.mock import patch
from unittest.mock import MagicMock, patch
import pytest
import torch
@@ -48,7 +47,7 @@ def _build_attention(mm_config, attn_backend=None):
invokes ``process_weights_after_loading`` to simulate the model loader's
auto-scan. Yields ``None`` if FlashInfer cuDNN is not available.
"""
from vllm.config import VllmConfig, set_current_vllm_config
from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config
from vllm.model_executor.layers.attention.mm_encoder_attention import (
MMEncoderAttention,
)
@@ -65,7 +64,7 @@ def _build_attention(mm_config, attn_backend=None):
return
vllm_config = VllmConfig()
vllm_config.model_config = SimpleNamespace(multimodal_config=mm_config)
vllm_config.model_config = MagicMock(spec=ModelConfig, multimodal_config=mm_config)
with (
set_current_vllm_config(vllm_config),
@@ -212,7 +211,7 @@ def test_aiter_static_scales_loaded(tmp_path) -> None:
def test_static_scales_missing_layer(tmp_path) -> None:
"""Verify error when requested layer is not in the scale file."""
from vllm.config import VllmConfig, set_current_vllm_config
from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config
from vllm.config.multimodal import MultiModalConfig
from vllm.v1.attention.backends.registry import AttentionBackendEnum
@@ -228,7 +227,7 @@ def test_static_scales_missing_layer(tmp_path) -> None:
mm_encoder_fp8_scale_path=str(scale_file),
)
vllm_config = VllmConfig()
vllm_config.model_config = SimpleNamespace(multimodal_config=mm_config)
vllm_config.model_config = MagicMock(spec=ModelConfig, multimodal_config=mm_config)
from vllm.model_executor.layers.attention.mm_encoder_attention import (
MMEncoderAttention,