[Misc] Cleanup: use consistent pytest-mock in unit tests (#2698)

Signed-off-by: yuanheng <jonathan.zhaoyh@gmail.com>
This commit is contained in:
Yuanheng Zhao authored and GitHub committed 2026-04-13 19:21:42 +08:00
1 parent 2c67c30550
commit e0cdbe9a5d
26 files changed
+1598 -1369

No files matched your search

@@ -10,10 +10,9 @@ Regression test for: https://github.com/vllm-project/vllm/issues/34506
- Interleaved (use_audio_in_video) should also work correctly.
"""
from unittest.mock import Mock
import pytest
import torch
from pytest_mock import MockerFixture
from vllm.model_executor.models.qwen2_5_omni_thinker import (
check_interleaved_audio_video,
merge_interleaved_embeddings,
@@ -107,7 +106,7 @@ class TestCheckInterleavedAudioVideo:
# ---------------------------------------------------------------------------
def make_mock_model(hidden: int = 8):
def make_mock_model(mocker: MockerFixture, hidden: int = 8):
"""
Return a minimal mock of Qwen2_5OmniThinkerForConditionalGeneration
that has enough structure to run embed_input_ids.
@@ -116,10 +115,10 @@ def make_mock_model(hidden: int = 8):
Qwen2_5OmniThinkerForConditionalGeneration,
)
model = Mock(spec=Qwen2_5OmniThinkerForConditionalGeneration)
model = mocker.Mock(spec=Qwen2_5OmniThinkerForConditionalGeneration)
# Config with token IDs
cfg = Mock()
cfg = mocker.Mock()
cfg.video_token_index = VIDEO_TOKEN_ID
cfg.audio_token_index = AUDIO_TOKEN_ID
model.config = cfg
@@ -130,9 +129,9 @@ def make_mock_model(hidden: int = 8):
# view with shared memory, which masked_scatter_ cannot handle).
return ids.float().unsqueeze(-1).expand(-1, hidden).clone()
lang_model = Mock()
lang_model = mocker.Mock()
lang_model.embed_input_ids = fake_lm_embed
model.get_language_model = Mock(return_value=lang_model)
model.get_language_model = mocker.Mock(return_value=lang_model)
from vllm.model_executor.models.interfaces import SupportsMultiModal
@@ -169,7 +168,7 @@ def build_mm_embeds(audio_n, image_n, video_n, hidden, audio_val=10.0, image_val
class TestEmbedInputIds:
def _run(self, audio_n, image_n, video_n, hidden=8):
def _run(self, mocker: MockerFixture, audio_n, image_n, video_n, hidden=8):
"""
Run embed_input_ids for a non-interleaved mixed-modality sequence.
Returns (result_embeds, input_ids, is_multimodal).
@@ -177,33 +176,33 @@ class TestEmbedInputIds:
input_ids, is_multimodal = make_token_seq(audio_n, image_n, video_n)
mm_embeds = build_mm_embeds(audio_n, image_n, video_n, hidden)
model, _ = make_mock_model(hidden)
model, _ = make_mock_model(mocker, hidden)
result = model.embed_input_ids(input_ids, mm_embeds, is_multimodal=is_multimodal)
return result, input_ids, is_multimodal
def test_audio_only(self):
def test_audio_only(self, mocker: MockerFixture):
"""Audio-only: audio positions get audio embeddings."""
audio_n, hidden = 5, 8
audio_val = 10.0
result, input_ids, is_multimodal = self._run(audio_n, 0, 0, hidden)
result, input_ids, is_multimodal = self._run(mocker, audio_n, 0, 0, hidden)
audio_pos = (input_ids == AUDIO_TOKEN_ID).nonzero(as_tuple=True)[0]
assert result[audio_pos].allclose(torch.full((audio_n, hidden), audio_val)), (
"Audio positions should get audio embeddings"
)
def test_video_only(self):
def test_video_only(self, mocker: MockerFixture):
"""Video-only: video positions get video embeddings."""
video_n, hidden = 6, 8
video_val = 30.0
result, input_ids, is_multimodal = self._run(0, 0, video_n, hidden)
result, input_ids, is_multimodal = self._run(mocker, 0, 0, video_n, hidden)
video_pos = (input_ids == VIDEO_TOKEN_ID).nonzero(as_tuple=True)[0]
assert result[video_pos].allclose(torch.full((video_n, hidden), video_val)), (
"Video positions should get video embeddings"
)
def test_mixed_modalities_audio_goes_to_audio_pos(self):
def test_mixed_modalities_audio_goes_to_audio_pos(self, mocker: MockerFixture):
"""
Regression test for GitHub issue #34506:
With audio + image + video (non-interleaved), audio positions must
@@ -212,7 +211,7 @@ class TestEmbedInputIds:
audio_n, image_n, video_n, hidden = 5, 4, 6, 8
audio_val, image_val, video_val = 10.0, 20.0, 30.0
result, input_ids, is_multimodal = self._run(audio_n, image_n, video_n, hidden)
result, input_ids, is_multimodal = self._run(mocker, audio_n, image_n, video_n, hidden)
audio_pos = (input_ids == AUDIO_TOKEN_ID).nonzero(as_tuple=True)[0]
image_pos = (input_ids == IMAGE_TOKEN_ID).nonzero(as_tuple=True)[0]
@@ -233,10 +232,10 @@ class TestEmbedInputIds:
f"Video emb wrong: expected {video_val}, got mean={mean_v:.1f}"
)
def test_text_positions_unchanged(self):
def test_text_positions_unchanged(self, mocker: MockerFixture):
"""Text positions should keep their text embeddings."""
audio_n, image_n, video_n, hidden = 3, 2, 4, 8
result, input_ids, is_multimodal = self._run(audio_n, image_n, video_n, hidden)
result, input_ids, is_multimodal = self._run(mocker, audio_n, image_n, video_n, hidden)
text_pos = (~is_multimodal).nonzero(as_tuple=True)[0]
# Text tokens have value TEXT_TOKEN_ID=0, so embed -> 0.0
@@ -244,7 +243,7 @@ class TestEmbedInputIds:
"Text positions should keep text embeddings"
)
def test_interleaved_use_audio_in_video(self):
def test_interleaved_use_audio_in_video(self, mocker: MockerFixture):
"""
Interleaved (use_audio_in_video): video chunks interleaved with audio.
Video embeddings must go to video positions, audio to audio positions.
@@ -263,7 +262,7 @@ class TestEmbedInputIds:
torch.full((audio_n, hidden), audio_val),
]
model, _ = make_mock_model(hidden)
model, _ = make_mock_model(mocker, hidden)
result = model.embed_input_ids(input_ids, mm_embeds, is_multimodal=is_multimodal)
video_pos = (input_ids == VIDEO_TOKEN_ID).nonzero(as_tuple=True)[0]