[Feat] Support Pipeline Resolution Based on HF Config & Make Pipeline Registry Eager (#3760)
Signed-off-by: Alex Brooks <albrooks@redhat.com>
This commit is contained in:
1 parent
f6894f4308
commit
465f1464d2
44 files changed
+1038
-1126
No files matched your search
@@ -1,146 +1,68 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
"""Tests for the central pipeline registry (2.5/N)."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import logging
|
||||
"""Tests for out of tree registration to OMNI_PIPELINES."""
|
||||
|
||||
import pytest
|
||||
from transformers import PretrainedConfig
|
||||
|
||||
from vllm_omni.config.pipeline_registry import _OMNI_PIPELINES
|
||||
from vllm_omni.config.stage_config import (
|
||||
_PIPELINE_REGISTRY,
|
||||
PipelineConfig,
|
||||
StageExecutionType,
|
||||
StagePipelineConfig,
|
||||
register_pipeline,
|
||||
)
|
||||
from vllm_omni.config.pipeline_registry import OMNI_PIPELINES, register_pipeline
|
||||
from vllm_omni.config.stage_config import PipelineConfig, pipeline_cfg_resolver
|
||||
|
||||
pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
|
||||
|
||||
|
||||
class TestCentralRegistryDeclarations:
|
||||
"""Every in-tree pipeline must be declared exactly once in the central registry."""
|
||||
@pytest.fixture
|
||||
def custom_resolver():
|
||||
"""Build a reusable custom resolver for PipelineConfigs."""
|
||||
|
||||
def test_omni_entries_visible_in_registry(self):
|
||||
for key in _OMNI_PIPELINES:
|
||||
assert key in _PIPELINE_REGISTRY
|
||||
class CustomConfigType(PretrainedConfig):
|
||||
pass
|
||||
|
||||
def test_expected_omni_pipelines_present(self):
|
||||
# Guard against accidental removal during future refactors.
|
||||
assert "qwen2_5_omni" in _OMNI_PIPELINES
|
||||
assert "qwen2_5_omni_thinker_only" in _OMNI_PIPELINES
|
||||
assert "qwen3_omni_moe" in _OMNI_PIPELINES
|
||||
assert "aura_omni" in _OMNI_PIPELINES
|
||||
assert "qwen3_tts" in _OMNI_PIPELINES
|
||||
@pipeline_cfg_resolver(config_type=CustomConfigType)
|
||||
def custom_resolver(
|
||||
hf_config: CustomConfigType,
|
||||
) -> PipelineConfig:
|
||||
return PipelineConfig(model_type="resolved_type")
|
||||
|
||||
return custom_resolver
|
||||
|
||||
|
||||
class TestLazyLoading:
|
||||
"""Pipelines are imported only on first access."""
|
||||
|
||||
def test_contains_without_import(self):
|
||||
# ``in`` hits the lazy map, not the loaded cache.
|
||||
assert "qwen3_omni_moe" in _PIPELINE_REGISTRY
|
||||
|
||||
def test_getitem_loads_correct_pipeline(self):
|
||||
pipeline = _PIPELINE_REGISTRY["qwen3_omni_moe"]
|
||||
assert pipeline.model_type == "qwen3_omni_moe"
|
||||
assert pipeline.model_arch == "Qwen3OmniMoeForConditionalGeneration"
|
||||
|
||||
def test_getitem_loads_aura_omni_pipeline(self):
|
||||
pipeline = _PIPELINE_REGISTRY["aura_omni"]
|
||||
assert pipeline.model_type == "aura_omni"
|
||||
assert [stage.model_stage for stage in pipeline.stages] == [
|
||||
"asr",
|
||||
"aura",
|
||||
"qwen3_tts",
|
||||
"code2wav",
|
||||
]
|
||||
assert pipeline.stages[0].model_arch == "Qwen3ASRForConditionalGeneration"
|
||||
assert pipeline.stages[1].model_arch == "AuraQwen3VLForConditionalGeneration"
|
||||
assert pipeline.stages[2].model_arch == "Qwen3TTSTalkerForConditionalGeneration"
|
||||
assert pipeline.stages[3].model_arch == "Qwen3TTSCode2Wav"
|
||||
|
||||
def test_unknown_model_type_returns_none_via_get(self):
|
||||
assert _PIPELINE_REGISTRY.get("not_a_real_pipeline") is None
|
||||
|
||||
def test_unknown_model_type_raises_keyerror_via_getitem(self):
|
||||
with pytest.raises(KeyError):
|
||||
_PIPELINE_REGISTRY["not_a_real_pipeline"]
|
||||
|
||||
def test_iteration_yields_registered_pipelines(self):
|
||||
keys = set(_PIPELINE_REGISTRY)
|
||||
assert "qwen2_5_omni" in keys
|
||||
assert "qwen3_omni_moe" in keys
|
||||
|
||||
def test_iteration_resilience(self, monkeypatch, caplog):
|
||||
# Synthetic failure to verify iteration robustness against load errors.
|
||||
target_logger = logging.getLogger("vllm_omni.config.stage_config")
|
||||
target_logger.addHandler(caplog.handler)
|
||||
prev_level = target_logger.level
|
||||
target_logger.setLevel(logging.ERROR)
|
||||
try:
|
||||
monkeypatch.setattr(
|
||||
_PIPELINE_REGISTRY,
|
||||
"_lazy_map",
|
||||
{
|
||||
"healthy": ("vllm_omni.model_executor.models.qwen2_5_omni.pipeline", "QWEN2_5_OMNI_PIPELINE"),
|
||||
"broken": ("vllm_omni.non_existent_module", "SomeConfig"),
|
||||
"attr_fail": ("vllm_omni.config.stage_config", "NON_EXISTENT_VAR"),
|
||||
},
|
||||
)
|
||||
monkeypatch.setattr(_PIPELINE_REGISTRY, "_loaded", {})
|
||||
|
||||
# Assert that list(_PIPELINE_REGISTRY.values()) and dict(_PIPELINE_REGISTRY.items())
|
||||
# both yield only healthy entries.
|
||||
healthy_instance = _PIPELINE_REGISTRY["healthy"]
|
||||
assert list(_PIPELINE_REGISTRY.values()) == [healthy_instance]
|
||||
assert dict(_PIPELINE_REGISTRY.items()) == {"healthy": healthy_instance}
|
||||
assert "Failed to import pipeline module" in caplog.text
|
||||
assert "Pipeline variable 'NON_EXISTENT_VAR' not found" in caplog.text
|
||||
|
||||
finally:
|
||||
target_logger.removeHandler(caplog.handler)
|
||||
target_logger.setLevel(prev_level)
|
||||
def test_register_pipeline_config(clean_pipeline_registry):
|
||||
"""Ensure that we can register a custom pipeline config to OMNI_PIPELINES."""
|
||||
new_model_type = "new_model_type"
|
||||
pipe_cfg = PipelineConfig(model_type=new_model_type)
|
||||
assert new_model_type not in OMNI_PIPELINES
|
||||
register_pipeline(pipe_cfg)
|
||||
assert new_model_type in OMNI_PIPELINES
|
||||
assert OMNI_PIPELINES[new_model_type] is pipe_cfg
|
||||
|
||||
|
||||
class TestDynamicRegistration:
|
||||
"""``register_pipeline()`` still works for plugins and tests."""
|
||||
def test_register_pipeline_config_with_model_type(clean_pipeline_registry):
|
||||
"""Ensure that we can register a custom pipeline config with an explicit model_type to OMNI_PIPELINES."""
|
||||
new_model_type = "new_model_type"
|
||||
unused_model_type = "foo"
|
||||
pipe_cfg = PipelineConfig(model_type=unused_model_type)
|
||||
assert new_model_type not in OMNI_PIPELINES
|
||||
assert unused_model_type not in OMNI_PIPELINES
|
||||
|
||||
def test_register_adds_to_registry(self):
|
||||
custom = PipelineConfig(
|
||||
model_type="_test_dynamic_registration",
|
||||
model_arch="DynamicTestModel",
|
||||
stages=(
|
||||
StagePipelineConfig(
|
||||
stage_id=0,
|
||||
model_stage="test",
|
||||
execution_type=StageExecutionType.LLM_AR,
|
||||
input_sources=(),
|
||||
final_output=True,
|
||||
),
|
||||
),
|
||||
)
|
||||
register_pipeline(custom)
|
||||
try:
|
||||
assert "_test_dynamic_registration" in _PIPELINE_REGISTRY
|
||||
assert _PIPELINE_REGISTRY["_test_dynamic_registration"] is custom
|
||||
finally:
|
||||
# Don't leak the test registration into other tests.
|
||||
if "_test_dynamic_registration" in _PIPELINE_REGISTRY:
|
||||
del _PIPELINE_REGISTRY["_test_dynamic_registration"]
|
||||
# Registering with an explicitly provided model_type uses
|
||||
# the passed value instead of the pipeline_cfg.model_type
|
||||
register_pipeline(pipe_cfg, new_model_type)
|
||||
assert new_model_type in OMNI_PIPELINES
|
||||
assert unused_model_type not in OMNI_PIPELINES
|
||||
assert OMNI_PIPELINES[new_model_type] is pipe_cfg
|
||||
|
||||
def test_dynamic_registration_overrides_lazy_entry(self):
|
||||
# Build a substitute for qwen3_omni_moe that we can distinguish.
|
||||
original = _PIPELINE_REGISTRY["qwen3_omni_moe"]
|
||||
override = PipelineConfig(
|
||||
model_type="qwen3_omni_moe",
|
||||
model_arch="OverriddenArch",
|
||||
stages=original.stages,
|
||||
)
|
||||
register_pipeline(override)
|
||||
try:
|
||||
assert _PIPELINE_REGISTRY["qwen3_omni_moe"].model_arch == "OverriddenArch"
|
||||
finally:
|
||||
# Remove the dynamic override so later tests see the original.
|
||||
if "qwen3_omni_moe" in _PIPELINE_REGISTRY._loaded:
|
||||
del _PIPELINE_REGISTRY["qwen3_omni_moe"]
|
||||
|
||||
def test_register_resolver(custom_resolver, clean_pipeline_registry):
|
||||
"""Ensure that we can register a custom resolver to OMNI_PIPELINES."""
|
||||
new_model_type = "new_model_type"
|
||||
assert new_model_type not in OMNI_PIPELINES
|
||||
register_pipeline(custom_resolver, new_model_type)
|
||||
assert new_model_type in OMNI_PIPELINES
|
||||
assert OMNI_PIPELINES[new_model_type] is custom_resolver
|
||||
|
||||
|
||||
def test_register_resolver_requires_model_type(custom_resolver, clean_pipeline_registry):
|
||||
"""Ensure that registering a custom resolver to OMNI_PIPELINES requires an explicit model_type."""
|
||||
with pytest.raises(ValueError):
|
||||
register_pipeline(custom_resolver)
|
||||
@@ -17,6 +17,7 @@ from tests.model_executor.helpers import bootstrap_vllm_layer_custom_op_modules
|
||||
bootstrap_vllm_layer_custom_op_modules()
|
||||
|
||||
pytest_plugins = (
|
||||
"tests.helpers.fixtures.config",
|
||||
"tests.helpers.fixtures.env",
|
||||
"tests.helpers.fixtures.log",
|
||||
"tests.helpers.fixtures.run_args",
|
||||
|
||||
@@ -14,7 +14,7 @@ from tests.helpers.stage_config import get_deploy_config_path, modify_stage_conf
|
||||
from vllm_omni.platforms import current_omni_platform
|
||||
|
||||
models = ["Qwen/Qwen3-Omni-30B-A3B-Instruct"]
|
||||
|
||||
thinker_only_models = ["Qwen/Qwen3-Omni-30B-A3B-Captioner"]
|
||||
|
||||
# Single CI deploy YAML; rocm/xpu deltas are picked automatically via the
|
||||
# platforms: section. Only CUDA needs an extra enforce_eager tweak.
|
||||
@@ -40,6 +40,10 @@ else:
|
||||
|
||||
# Create parameter combinations for model and stage config
|
||||
test_params = [(model, stage_config) for model in models for stage_config in stage_configs]
|
||||
# we can use the same config for a model that only has thinker (i.e., does not
|
||||
# enable audio output) because the resolver should figure out that it doesn't
|
||||
# need the full pipeline based on the HF config.
|
||||
thinker_test_params = [(model, stage_config) for model in thinker_only_models for stage_config in stage_configs]
|
||||
|
||||
|
||||
def get_question(prompt_type="video"):
|
||||
@@ -61,3 +65,15 @@ def test_video_to_audio(omni_runner, omni_runner_handler) -> None:
|
||||
|
||||
# Test single completion
|
||||
omni_runner_handler.send_omni_request(request_config)
|
||||
|
||||
|
||||
@pytest.mark.advanced_model
|
||||
@pytest.mark.omni
|
||||
@hardware_test(res={"cuda": "H100", "rocm": "MI325"}, num_cards=1)
|
||||
@pytest.mark.parametrize("omni_runner", thinker_test_params, indirect=True)
|
||||
def test_thinker_only_model_request(omni_runner, omni_runner_handler) -> None:
|
||||
"""Test that we can load and run a request through a model that only has the thinker stage."""
|
||||
request_config = {"prompts": "what color is the sky?", "modalities": ["text"]}
|
||||
|
||||
# Test single completion
|
||||
omni_runner_handler.send_omni_request(request_config)
|
||||
@@ -9,13 +9,13 @@ from vllm.inputs import PromptType
|
||||
|
||||
from tests.helpers.mark import hardware_test
|
||||
from tests.helpers.stage_config import get_deploy_config_path
|
||||
from vllm_omni.config.stage_config import _PIPELINE_REGISTRY
|
||||
from vllm_omni.config.pipeline_registry import OMNI_PIPELINES
|
||||
from vllm_omni.entrypoints.async_omni import AsyncOmni
|
||||
|
||||
os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
|
||||
|
||||
SEED = 42
|
||||
pipeline = _PIPELINE_REGISTRY["qwen2_5_omni_thinker_only"]
|
||||
pipeline = OMNI_PIPELINES["qwen2_5_omni_thinker_only"]
|
||||
|
||||
# Single-stage thinker-only deploy, materialized from tests.helpers.stage_config._CI_OVERLAYS.
|
||||
stage_config = get_deploy_config_path("ci/qwen2_5_omni_thinker_only.yaml")
|
||||
|
||||
@@ -24,7 +24,7 @@ def test_dreamzero_vla_resolves_to_dreamzero_config(monkeypatch):
|
||||
|
||||
def test_dreamzero_config_sets_model_class_and_policy_config(monkeypatch):
|
||||
monkeypatch.setattr(
|
||||
"vllm_omni.config.stage_config.StageConfigFactory._auto_detect_model_type",
|
||||
"vllm_omni.config.config_factory.StageConfigFactory._auto_detect_model_type",
|
||||
classmethod(lambda _cls, _model, trust_remote_code=True: ("vla", None)),
|
||||
)
|
||||
monkeypatch.setattr(
|
||||
|
||||
@@ -0,0 +1,14 @@
|
||||
"""Fixtures config creation & out of tree registry management."""
|
||||
|
||||
import pytest
|
||||
|
||||
from vllm_omni.config.pipeline_registry import OMNI_PIPELINES
|
||||
|
||||
|
||||
@pytest.fixture
|
||||
def clean_pipeline_registry():
|
||||
"""Ensure the OMNI_PIPELINES are in a clean state for a test that mutates it."""
|
||||
snapshot = dict(OMNI_PIPELINES)
|
||||
yield
|
||||
OMNI_PIPELINES.clear()
|
||||
OMNI_PIPELINES.update(snapshot)
|
||||
@@ -1,76 +0,0 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
"""Unit tests for GLM-Image model __init__.py lazy import pattern.
|
||||
|
||||
The ``__init__.py`` uses ``__getattr__`` for lazy loading to avoid importing
|
||||
``transformers.models.glm_image`` at module init, which may not be available
|
||||
in all environments.
|
||||
"""
|
||||
|
||||
import pytest
|
||||
|
||||
pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
|
||||
|
||||
|
||||
class TestLazyImport:
|
||||
"""Test the __getattr__ lazy import pattern in glm_image __init__.py."""
|
||||
|
||||
def test_getattr_exists_and_is_callable(self):
|
||||
"""Verify __getattr__ exists and is callable."""
|
||||
import vllm_omni.model_executor.models.glm_image as glm_image_pkg
|
||||
|
||||
assert hasattr(glm_image_pkg, "__getattr__")
|
||||
assert callable(glm_image_pkg.__getattr__)
|
||||
|
||||
def test_getattr_returns_class_for_known_attribute(self):
|
||||
"""Verify __getattr__ returns GlmImageForConditionalGeneration for known attribute."""
|
||||
import vllm_omni.model_executor.models.glm_image as glm_image_pkg
|
||||
|
||||
# Call __getattr__ directly to test the lazy import logic
|
||||
result = glm_image_pkg.__getattr__("GlmImageForConditionalGeneration")
|
||||
|
||||
# Verify we get a class (the actual GlmImageForConditionalGeneration)
|
||||
assert result is not None
|
||||
assert isinstance(result, type)
|
||||
assert result.__name__ == "GlmImageForConditionalGeneration"
|
||||
|
||||
def test_getattr_raises_for_unknown_attribute(self):
|
||||
"""Verify __getattr__ raises AttributeError for unknown attributes."""
|
||||
import vllm_omni.model_executor.models.glm_image as glm_image_pkg
|
||||
|
||||
# Test unknown attribute via __getattr__ directly
|
||||
with pytest.raises(AttributeError, match="has no attribute"):
|
||||
glm_image_pkg.__getattr__("UnknownClass")
|
||||
|
||||
def test___all___exports_correct_symbols(self):
|
||||
"""Verify __all__ contains the expected exported symbols."""
|
||||
import vllm_omni.model_executor.models.glm_image as glm_image_pkg
|
||||
|
||||
assert hasattr(glm_image_pkg, "__all__")
|
||||
assert "GlmImageForConditionalGeneration" in glm_image_pkg.__all__
|
||||
|
||||
|
||||
class TestLazyImportDoesNotImportTransformersAtInit:
|
||||
"""Verify that importing the package does not eagerly load transformers."""
|
||||
|
||||
def test_glm_image_module_has_getattr(self):
|
||||
"""Test that the module has __getattr__ for lazy loading."""
|
||||
import vllm_omni.model_executor.models.glm_image as glm_image_pkg
|
||||
|
||||
# The module should have __getattr__
|
||||
assert hasattr(glm_image_pkg, "__getattr__")
|
||||
assert callable(glm_image_pkg.__getattr__)
|
||||
|
||||
# And __all__
|
||||
assert hasattr(glm_image_pkg, "__all__")
|
||||
|
||||
def test_module_has_proper_structure(self):
|
||||
"""Test that the module has proper Python module structure."""
|
||||
import vllm_omni.model_executor.models.glm_image as glm_image_pkg
|
||||
|
||||
# Should be a module
|
||||
assert hasattr(glm_image_pkg, "__name__")
|
||||
assert glm_image_pkg.__name__ == "vllm_omni.model_executor.models.glm_image"
|
||||
|
||||
# Should have __file__ (even if it's a package)
|
||||
assert hasattr(glm_image_pkg, "__file__")
|
||||
@@ -19,9 +19,8 @@ from types import SimpleNamespace
|
||||
|
||||
import pytest
|
||||
|
||||
from vllm_omni.config.pipeline_registry import _OMNI_PIPELINES
|
||||
from vllm_omni.config.pipeline_registry import OMNI_PIPELINES
|
||||
from vllm_omni.config.stage_config import (
|
||||
_PIPELINE_REGISTRY,
|
||||
PipelineConfig,
|
||||
StageExecutionType,
|
||||
)
|
||||
@@ -34,13 +33,13 @@ _PIPELINE_KEY = "minicpmo_4_5"
|
||||
|
||||
class TestRegistryDeclaration:
|
||||
def test_declared_in_omni_pipelines(self) -> None:
|
||||
assert _PIPELINE_KEY in _OMNI_PIPELINES
|
||||
assert _PIPELINE_KEY in OMNI_PIPELINES
|
||||
|
||||
def test_visible_in_central_registry(self) -> None:
|
||||
assert _PIPELINE_KEY in _PIPELINE_REGISTRY
|
||||
assert _PIPELINE_KEY in OMNI_PIPELINES
|
||||
|
||||
def test_lazy_load_returns_pipeline_config(self) -> None:
|
||||
pipeline = _PIPELINE_REGISTRY[_PIPELINE_KEY]
|
||||
pipeline = OMNI_PIPELINES[_PIPELINE_KEY]
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
assert pipeline.model_type == _PIPELINE_KEY
|
||||
assert pipeline.model_arch == "MiniCPMO45OmniForConditionalGeneration"
|
||||
@@ -49,7 +48,7 @@ class TestRegistryDeclaration:
|
||||
class TestPipelineTopology:
|
||||
@pytest.fixture(scope="class")
|
||||
def pipeline(self) -> PipelineConfig:
|
||||
return _PIPELINE_REGISTRY[_PIPELINE_KEY]
|
||||
return OMNI_PIPELINES[_PIPELINE_KEY]
|
||||
|
||||
def test_two_stages(self, pipeline: PipelineConfig) -> None:
|
||||
assert len(pipeline.stages) == 2
|
||||
@@ -99,7 +98,7 @@ class TestArchAliases:
|
||||
|
||||
@pytest.fixture(scope="class")
|
||||
def pipeline(self) -> PipelineConfig:
|
||||
return _PIPELINE_REGISTRY[_PIPELINE_KEY]
|
||||
return OMNI_PIPELINES[_PIPELINE_KEY]
|
||||
|
||||
def test_shared_minicpmo_alias_present(self, pipeline: PipelineConfig) -> None:
|
||||
# MiniCPM-o 4.5 ships ``architectures=["MiniCPMO"]`` in its HF config.
|
||||
@@ -123,7 +122,7 @@ class TestHfConfigPredicate:
|
||||
|
||||
@pytest.fixture(scope="class")
|
||||
def predicate(self):
|
||||
pipeline = _PIPELINE_REGISTRY[_PIPELINE_KEY]
|
||||
pipeline = OMNI_PIPELINES[_PIPELINE_KEY]
|
||||
assert pipeline.hf_config_predicate is not None, (
|
||||
"MiniCPM-o 4.5 pipeline must declare hf_config_predicate to "
|
||||
"avoid misrouting MiniCPM-o 2.6 checkpoints into the 4.5 path."
|
||||
|
||||
+252
-73
@@ -12,14 +12,15 @@ from pathlib import Path
|
||||
from unittest.mock import patch
|
||||
|
||||
import pytest
|
||||
from transformers import PretrainedConfig, Qwen3OmniMoeConfig
|
||||
|
||||
from tests.helpers.stage_config import get_deploy_config_path
|
||||
from vllm_omni.config.config_factory import StageConfigFactory
|
||||
from vllm_omni.config.pipeline_registry import OMNI_PIPELINES, register_pipeline
|
||||
from vllm_omni.config.stage_config import (
|
||||
_PIPELINE_REGISTRY,
|
||||
DeployConfig,
|
||||
PipelineConfig,
|
||||
StageConfig,
|
||||
StageConfigFactory,
|
||||
StageDeployConfig,
|
||||
StageExecutionType,
|
||||
StagePipelineConfig,
|
||||
@@ -30,14 +31,16 @@ from vllm_omni.config.stage_config import (
|
||||
build_stage_runtime_overrides,
|
||||
load_deploy_config,
|
||||
merge_pipeline_deploy,
|
||||
register_pipeline,
|
||||
pipeline_cfg_resolver,
|
||||
strip_parent_engine_args,
|
||||
)
|
||||
from vllm_omni.engine.arg_utils import SHARED_FIELDS, EngineArgs, internal_blacklist_keys
|
||||
|
||||
pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
|
||||
|
||||
pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
|
||||
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG = Qwen3OmniMoeConfig(enable_audio_output=True)
|
||||
Q3_OMNI_THINKER_HF_CONFIG = Qwen3OmniMoeConfig(enable_audio_output=False)
|
||||
|
||||
|
||||
class TestStageType:
|
||||
@@ -511,26 +514,39 @@ class TestStageResolutionHelpers:
|
||||
|
||||
|
||||
class TestPipelineDiscovery:
|
||||
"""Tests for the central pipeline registry (``pipeline_registry._OMNI_PIPELINES``)."""
|
||||
"""Tests for the central pipeline registry (``OMNI_PIPELINES``)."""
|
||||
|
||||
def test_registry_has_known_models(self):
|
||||
"""Built-in pipelines are lazy-loaded from the central declaration
|
||||
on first access; no eager import or discovery walk needed."""
|
||||
# ``in`` triggers the lazy-map lookup without forcing a load.
|
||||
assert "qwen2_5_omni" in _PIPELINE_REGISTRY
|
||||
assert "qwen3_omni_moe" in _PIPELINE_REGISTRY
|
||||
assert "qwen3_tts" in _PIPELINE_REGISTRY
|
||||
"""Check that specific models are in OMNI_PIPELINES."""
|
||||
assert "qwen2_5_omni" in OMNI_PIPELINES
|
||||
assert "qwen3_omni_moe" in OMNI_PIPELINES
|
||||
assert "qwen3_tts" in OMNI_PIPELINES
|
||||
|
||||
def test_registry_loads_pipeline_on_getitem(self):
|
||||
"""Looking up a registered model_type returns the matching PipelineConfig."""
|
||||
pipeline = _PIPELINE_REGISTRY["qwen3_omni_moe"]
|
||||
def test_registry_resolver_qwen3_omni_all_stages(self):
|
||||
"""Test that providing the HF config for qwen3 omni with audio enabled uses all stages."""
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
assert pipeline.model_type == "qwen3_omni_moe"
|
||||
assert len(pipeline.stages) == 3 # thinker + talker + code2wav
|
||||
|
||||
def test_registry_resolver_qwen3_omni_thinker_only(self):
|
||||
"""Test that providing the HF config for qwen3 omni without audio is thinker only."""
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_THINKER_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
assert pipeline.model_type == "qwen3_omni_moe_thinker_only"
|
||||
assert len(pipeline.stages) == 1 # thinker only
|
||||
|
||||
def test_registry_returns_none_for_unknown(self):
|
||||
"""Unknown model_types aren't found; ``get()`` returns None."""
|
||||
assert "definitely_not_a_real_model" not in _PIPELINE_REGISTRY
|
||||
assert _PIPELINE_REGISTRY.get("definitely_not_a_real_model") is None
|
||||
"""Unknown model_types aren't found and resolve to `None`."""
|
||||
assert "definitely_not_a_real_model" not in OMNI_PIPELINES
|
||||
assert OMNI_PIPELINES.get("definitely_not_a_real_model") is None
|
||||
assert StageConfigFactory.resolve_pipeline_config("definitely_not_a_real_model") is None
|
||||
|
||||
def test_pipeline_config_supports_hf_architectures(self):
|
||||
"""PipelineConfig accepts hf_architectures for HF-arch fallback
|
||||
@@ -579,6 +595,67 @@ class TestPipelineConfigNew:
|
||||
assert any("no stages" in e.lower() for e in p.validate())
|
||||
|
||||
|
||||
class TestPipelineRegistration:
|
||||
def test_pipeline_registration(self, clean_pipeline_registry):
|
||||
"""Ensure that we can register and create a custom pipeline config."""
|
||||
new_model_type = "new_model_type"
|
||||
pipe_cfg = PipelineConfig(model_type=new_model_type)
|
||||
|
||||
# Register the new PipelineConfig
|
||||
assert new_model_type not in OMNI_PIPELINES
|
||||
register_pipeline(pipe_cfg)
|
||||
assert new_model_type in OMNI_PIPELINES
|
||||
assert OMNI_PIPELINES[new_model_type] is pipe_cfg
|
||||
|
||||
class FakeConfig(PretrainedConfig):
|
||||
model_type = new_model_type
|
||||
|
||||
# Create the model
|
||||
with (
|
||||
patch("vllm_omni.config.config_factory.get_config", return_value=FakeConfig()),
|
||||
patch.object(StageConfigFactory, "_create_from_registry") as mock_create,
|
||||
):
|
||||
StageConfigFactory.create_from_model("fake/model")
|
||||
mock_create.assert_called_once()
|
||||
# Ensure that the passed pipeline_config is the right type
|
||||
pipeline_cfg = mock_create.call_args[0][1]
|
||||
assert isinstance(pipeline_cfg, PipelineConfig)
|
||||
assert pipe_cfg.model_type == new_model_type
|
||||
|
||||
def test_resolver_registration(self, clean_pipeline_registry):
|
||||
"""Ensure that we can register and create a custom resolver."""
|
||||
new_model_type = "new_model_type"
|
||||
resolved_type = "resolved_type"
|
||||
|
||||
class FakeConfig(PretrainedConfig):
|
||||
model_type = new_model_type
|
||||
|
||||
@pipeline_cfg_resolver(config_type=FakeConfig)
|
||||
def custom_resolver(
|
||||
hf_config: FakeConfig,
|
||||
) -> PipelineConfig:
|
||||
return PipelineConfig(model_type=resolved_type)
|
||||
|
||||
# Register the new PipelineConfig
|
||||
assert new_model_type not in OMNI_PIPELINES
|
||||
|
||||
register_pipeline(custom_resolver, model_type=new_model_type)
|
||||
assert new_model_type in OMNI_PIPELINES
|
||||
assert OMNI_PIPELINES[new_model_type] is custom_resolver
|
||||
|
||||
# Create the model
|
||||
with (
|
||||
patch("vllm_omni.config.config_factory.get_config", return_value=FakeConfig()),
|
||||
patch.object(StageConfigFactory, "_create_from_registry") as mock_create,
|
||||
):
|
||||
StageConfigFactory.create_from_model("fake/model")
|
||||
mock_create.assert_called_once()
|
||||
# Ensure that the passed pipeline_config is the right type
|
||||
pipeline_cfg = mock_create.call_args[0][1]
|
||||
assert isinstance(pipeline_cfg, PipelineConfig)
|
||||
assert pipeline_cfg.model_type == resolved_type
|
||||
|
||||
|
||||
class TestResolveScheduler:
|
||||
def test_all_execution_types_handled(self):
|
||||
for et in StageExecutionType:
|
||||
@@ -603,18 +680,6 @@ class TestResolveScheduler:
|
||||
assert _resolve_scheduler(StageExecutionType.DIFFUSION) is None
|
||||
|
||||
|
||||
class TestPipelineRegistry:
|
||||
def test_register_and_lookup(self):
|
||||
p = PipelineConfig(
|
||||
model_type="__test_only__",
|
||||
model_arch="A",
|
||||
stages=(StagePipelineConfig(stage_id=0, model_stage="a"),),
|
||||
)
|
||||
register_pipeline(p)
|
||||
assert _PIPELINE_REGISTRY["__test_only__"] is p
|
||||
del _PIPELINE_REGISTRY["__test_only__"]
|
||||
|
||||
|
||||
class TestDeployConfigLoading:
|
||||
def test_custom_voice_dir_is_pipeline_wide_engine_arg(self, tmp_path):
|
||||
deploy_path = tmp_path / "qwen3_tts_custom_voice.yaml"
|
||||
@@ -633,8 +698,9 @@ stages:
|
||||
)
|
||||
|
||||
deploy = load_deploy_config(deploy_path)
|
||||
pipeline = _PIPELINE_REGISTRY["qwen3_tts"]
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config("qwen3_tts")
|
||||
stages = merge_pipeline_deploy(pipeline, deploy)
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
|
||||
assert deploy.custom_voice_dir == str(custom_voice_dir)
|
||||
assert {s.yaml_engine_args.get("custom_voice_dir") for s in stages} == {str(custom_voice_dir)}
|
||||
@@ -664,11 +730,16 @@ stages:
|
||||
assert "compilation_config" not in deploy.stages[0].engine_extras
|
||||
|
||||
def test_merge_pipeline_deploy(self):
|
||||
pipeline = _PIPELINE_REGISTRY["qwen3_omni_moe"]
|
||||
deploy_path = Path(__file__).parent.parent / "vllm_omni" / "deploy" / "qwen3_omni_moe.yaml"
|
||||
if not deploy_path.exists():
|
||||
pytest.skip("Deploy config not found")
|
||||
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
|
||||
deploy = load_deploy_config(deploy_path)
|
||||
stages = merge_pipeline_deploy(pipeline, deploy)
|
||||
|
||||
@@ -680,7 +751,12 @@ stages:
|
||||
assert s0.yaml_extras["default_sampling_params"]["detokenize"] is True
|
||||
|
||||
def test_merge_pipeline_deploy_preserves_num_replicas(self, tmp_path):
|
||||
pipeline = _PIPELINE_REGISTRY["qwen3_omni_moe"]
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
|
||||
base = Path(__file__).parent.parent / "vllm_omni" / "deploy" / "qwen3_omni_moe.yaml"
|
||||
if not base.exists():
|
||||
pytest.skip("Deploy config not found")
|
||||
@@ -894,14 +970,24 @@ stages:
|
||||
|
||||
class TestQwen3OmniPipeline:
|
||||
def test_registered(self):
|
||||
p = _PIPELINE_REGISTRY.get("qwen3_omni_moe")
|
||||
assert p is not None
|
||||
p = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(p, PipelineConfig)
|
||||
assert p.model_arch == "Qwen3OmniMoeForConditionalGeneration"
|
||||
assert len(p.stages) == 3
|
||||
assert p.validate() == []
|
||||
|
||||
def test_thinker(self):
|
||||
s = _PIPELINE_REGISTRY["qwen3_omni_moe"].get_stage(0)
|
||||
p = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(0)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.model_stage == "thinker"
|
||||
assert s.execution_type == StageExecutionType.LLM_AR
|
||||
assert s.owns_tokenizer is True
|
||||
@@ -909,14 +995,28 @@ class TestQwen3OmniPipeline:
|
||||
assert s.sampling_constraints["detokenize"] is True
|
||||
|
||||
def test_talker(self):
|
||||
s = _PIPELINE_REGISTRY["qwen3_omni_moe"].get_stage(1)
|
||||
p = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(1)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.input_sources == (0,)
|
||||
assert s.sampling_constraints["stop_token_ids"] == [2150]
|
||||
assert s.custom_process_input_func is not None
|
||||
assert s.custom_process_next_stage_input_func is not None
|
||||
|
||||
def test_code2wav(self):
|
||||
s = _PIPELINE_REGISTRY["qwen3_omni_moe"].get_stage(2)
|
||||
p = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(2)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.execution_type == StageExecutionType.LLM_GENERATION
|
||||
assert s.final_output_type == "audio"
|
||||
assert s.custom_process_input_func is not None
|
||||
@@ -924,14 +1024,18 @@ class TestQwen3OmniPipeline:
|
||||
|
||||
class TestQwen2_5OmniPipeline:
|
||||
def test_registered(self):
|
||||
p = _PIPELINE_REGISTRY.get("qwen2_5_omni")
|
||||
assert p is not None
|
||||
p = StageConfigFactory.resolve_pipeline_config("qwen2_5_omni")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
assert p.model_arch == "Qwen2_5OmniForConditionalGeneration"
|
||||
assert len(p.stages) == 3
|
||||
assert p.validate() == []
|
||||
|
||||
def test_thinker(self):
|
||||
s = _PIPELINE_REGISTRY["qwen2_5_omni"].get_stage(0)
|
||||
p = StageConfigFactory.resolve_pipeline_config("qwen2_5_omni")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(0)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.model_stage == "thinker"
|
||||
assert s.execution_type == StageExecutionType.LLM_AR
|
||||
assert s.owns_tokenizer is True
|
||||
@@ -939,13 +1043,21 @@ class TestQwen2_5OmniPipeline:
|
||||
assert s.requires_multimodal_data is True
|
||||
|
||||
def test_talker(self):
|
||||
s = _PIPELINE_REGISTRY["qwen2_5_omni"].get_stage(1)
|
||||
p = StageConfigFactory.resolve_pipeline_config("qwen2_5_omni")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(1)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.input_sources == (0,)
|
||||
assert s.sampling_constraints["stop_token_ids"] == [8294]
|
||||
assert s.custom_process_input_func is not None
|
||||
|
||||
def test_code2wav(self):
|
||||
s = _PIPELINE_REGISTRY["qwen2_5_omni"].get_stage(2)
|
||||
p = StageConfigFactory.resolve_pipeline_config("qwen2_5_omni")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(2)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.execution_type == StageExecutionType.LLM_GENERATION
|
||||
assert s.final_output_type == "audio"
|
||||
assert s.engine_output_type == "audio"
|
||||
@@ -953,14 +1065,19 @@ class TestQwen2_5OmniPipeline:
|
||||
|
||||
class TestQwen3TTSPipeline:
|
||||
def test_registered(self):
|
||||
p = _PIPELINE_REGISTRY.get("qwen3_tts")
|
||||
p = StageConfigFactory.resolve_pipeline_config("qwen3_tts")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
assert p is not None
|
||||
assert p.model_arch == "Qwen3TTSTalkerForConditionalGeneration"
|
||||
assert len(p.stages) == 2
|
||||
assert p.validate() == []
|
||||
|
||||
def test_talker_stage(self):
|
||||
s = _PIPELINE_REGISTRY["qwen3_tts"].get_stage(0)
|
||||
p = StageConfigFactory.resolve_pipeline_config("qwen3_tts")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(0)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.model_stage == "qwen3_tts"
|
||||
assert s.execution_type == StageExecutionType.LLM_AR
|
||||
assert s.owns_tokenizer is True
|
||||
@@ -970,7 +1087,11 @@ class TestQwen3TTSPipeline:
|
||||
assert s.model_arch is None
|
||||
|
||||
def test_code2wav_stage_has_per_stage_model_arch(self):
|
||||
s = _PIPELINE_REGISTRY["qwen3_tts"].get_stage(1)
|
||||
p = StageConfigFactory.resolve_pipeline_config("qwen3_tts")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(1)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.execution_type == StageExecutionType.LLM_GENERATION
|
||||
assert s.final_output_type == "audio"
|
||||
assert s.engine_output_type == "audio"
|
||||
@@ -986,7 +1107,7 @@ class TestQwen3TTSPipeline:
|
||||
pytest.skip("qwen3_tts deploy yaml not found")
|
||||
|
||||
deploy = load_deploy_config(deploy_path)
|
||||
pipeline = _PIPELINE_REGISTRY["qwen3_tts"]
|
||||
pipeline = OMNI_PIPELINES["qwen3_tts"]
|
||||
stages = merge_pipeline_deploy(pipeline, deploy)
|
||||
|
||||
# Stage 0 inherits pipeline-level model_arch
|
||||
@@ -1025,14 +1146,18 @@ class TestQwen3TTSPipeline:
|
||||
|
||||
class TestMingFlashOmniPipeline:
|
||||
def test_registered(self):
|
||||
p = _PIPELINE_REGISTRY.get("ming_flash_omni")
|
||||
assert p is not None
|
||||
p = StageConfigFactory.resolve_pipeline_config("ming_flash_omni")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
assert p.model_arch == "MingFlashOmniForConditionalGeneration"
|
||||
assert len(p.stages) == 2
|
||||
assert p.validate() == []
|
||||
|
||||
def test_thinker_stage(self):
|
||||
s = _PIPELINE_REGISTRY["ming_flash_omni"].get_stage(0)
|
||||
p = StageConfigFactory.resolve_pipeline_config("ming_flash_omni")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(0)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.model_stage == "thinker"
|
||||
assert s.execution_type == StageExecutionType.LLM_AR
|
||||
assert s.owns_tokenizer is True
|
||||
@@ -1042,7 +1167,11 @@ class TestMingFlashOmniPipeline:
|
||||
assert s.sampling_constraints["detokenize"] is True
|
||||
|
||||
def test_talker_stage(self):
|
||||
s = _PIPELINE_REGISTRY["ming_flash_omni"].get_stage(1)
|
||||
p = StageConfigFactory.resolve_pipeline_config("ming_flash_omni")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(1)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.model_stage == "ming_tts"
|
||||
assert s.execution_type == StageExecutionType.LLM_GENERATION
|
||||
assert s.input_sources == (0,)
|
||||
@@ -1060,20 +1189,28 @@ class TestMingFlashOmniPipeline:
|
||||
Lazy string references only fail at first inference otherwise — this
|
||||
catches typos in the pipeline declaration at import / registration time.
|
||||
"""
|
||||
s = _PIPELINE_REGISTRY["ming_flash_omni"].get_stage(1)
|
||||
p = StageConfigFactory.resolve_pipeline_config("ming_flash_omni")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(1)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
module_path, _, attr = s.custom_process_input_func.rpartition(".")
|
||||
module = importlib.import_module(module_path)
|
||||
assert callable(getattr(module, attr))
|
||||
|
||||
def test_tts_pipeline_registered(self):
|
||||
p = _PIPELINE_REGISTRY.get("ming_flash_omni_tts")
|
||||
assert p is not None
|
||||
p = StageConfigFactory.resolve_pipeline_config("ming_flash_omni_tts")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
assert p.model_arch == "MingFlashOmniTalkerForConditionalGeneration"
|
||||
assert len(p.stages) == 1
|
||||
assert p.validate() == []
|
||||
|
||||
def test_tts_stage(self):
|
||||
s = _PIPELINE_REGISTRY["ming_flash_omni_tts"].get_stage(0)
|
||||
p = StageConfigFactory.resolve_pipeline_config("ming_flash_omni_tts")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(0)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.model_stage == "ming_tts"
|
||||
assert s.execution_type == StageExecutionType.LLM_GENERATION
|
||||
assert s.input_sources == ()
|
||||
@@ -1096,7 +1233,8 @@ class TestMingFlashOmniPipeline:
|
||||
# We won't test stage 0/1 colocation contract here,
|
||||
# as there could exist more variant of custom device setup
|
||||
|
||||
pipeline = _PIPELINE_REGISTRY["ming_flash_omni"]
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config("ming_flash_omni")
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
stages = merge_pipeline_deploy(pipeline, deploy)
|
||||
assert len(stages) == 2
|
||||
assert stages[0].yaml_engine_args["model_arch"] == "MingFlashOmniForConditionalGeneration"
|
||||
@@ -1112,20 +1250,24 @@ class TestMingFlashOmniPipeline:
|
||||
assert len(deploy.stages) == 1
|
||||
assert deploy.pipeline == "ming_flash_omni_tts"
|
||||
|
||||
pipeline = _PIPELINE_REGISTRY["ming_flash_omni_tts"]
|
||||
pipeline = OMNI_PIPELINES["ming_flash_omni_tts"]
|
||||
stages = merge_pipeline_deploy(pipeline, deploy)
|
||||
assert len(stages) == 1
|
||||
assert stages[0].yaml_engine_args["model_arch"] == "MingFlashOmniTalkerForConditionalGeneration"
|
||||
|
||||
def test_thinker_only_pipeline_registered(self):
|
||||
p = _PIPELINE_REGISTRY.get("ming_flash_omni_thinker_only")
|
||||
assert p is not None
|
||||
p = StageConfigFactory.resolve_pipeline_config("ming_flash_omni_thinker_only")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
assert p.model_arch == "MingFlashOmniForConditionalGeneration"
|
||||
assert len(p.stages) == 1
|
||||
assert p.validate() == []
|
||||
|
||||
def test_thinker_only_stage(self):
|
||||
s = _PIPELINE_REGISTRY["ming_flash_omni_thinker_only"].get_stage(0)
|
||||
p = StageConfigFactory.resolve_pipeline_config("ming_flash_omni_thinker_only")
|
||||
assert isinstance(p, PipelineConfig)
|
||||
|
||||
s = p.get_stage(0)
|
||||
assert isinstance(s, StagePipelineConfig)
|
||||
assert s.model_stage == "thinker"
|
||||
assert s.execution_type == StageExecutionType.LLM_AR
|
||||
assert s.input_sources == ()
|
||||
@@ -1146,20 +1288,21 @@ class TestMingFlashOmniPipeline:
|
||||
assert len(deploy.stages) == 1
|
||||
assert deploy.pipeline == "ming_flash_omni_thinker_only"
|
||||
|
||||
pipeline = _PIPELINE_REGISTRY["ming_flash_omni_thinker_only"]
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config("ming_flash_omni_thinker_only")
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
stages = merge_pipeline_deploy(pipeline, deploy)
|
||||
assert len(stages) == 1
|
||||
assert stages[0].yaml_engine_args["model_arch"] == "MingFlashOmniForConditionalGeneration"
|
||||
|
||||
def test_image_pipeline_registered(self):
|
||||
p = _PIPELINE_REGISTRY.get("ming_flash_omni_image")
|
||||
p = OMNI_PIPELINES.get("ming_flash_omni_image")
|
||||
assert p is not None
|
||||
assert p.model_arch == "MingFlashOmniForConditionalGeneration"
|
||||
assert len(p.stages) == 2
|
||||
assert p.validate() == []
|
||||
|
||||
def test_image_thinker_stage(self):
|
||||
s = _PIPELINE_REGISTRY["ming_flash_omni_image"].get_stage(0)
|
||||
s = StageConfigFactory.resolve_pipeline_config("ming_flash_omni_image").get_stage(0)
|
||||
assert s.model_stage == "thinker"
|
||||
assert s.execution_type == StageExecutionType.LLM_AR
|
||||
assert s.input_sources == ()
|
||||
@@ -1173,7 +1316,7 @@ class TestMingFlashOmniPipeline:
|
||||
assert s.prompt_expand_func is not None
|
||||
|
||||
def test_image_dit_stage(self):
|
||||
s = _PIPELINE_REGISTRY["ming_flash_omni_image"].get_stage(1)
|
||||
s = StageConfigFactory.resolve_pipeline_config("ming_flash_omni_image").get_stage(1)
|
||||
assert s.model_stage == "dit"
|
||||
assert s.execution_type == StageExecutionType.DIFFUSION
|
||||
assert s.input_sources == (0,)
|
||||
@@ -1185,8 +1328,11 @@ class TestMingFlashOmniPipeline:
|
||||
|
||||
def test_image_processor_wiring_resolves(self):
|
||||
"""The prompt_expand_func and custom_process_input_func strings must point to real callables."""
|
||||
thinker = _PIPELINE_REGISTRY["ming_flash_omni_image"].get_stage(0)
|
||||
dit = _PIPELINE_REGISTRY["ming_flash_omni_image"].get_stage(1)
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config("ming_flash_omni_image")
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
|
||||
thinker = pipeline.get_stage(0)
|
||||
dit = pipeline.get_stage(1)
|
||||
for ref in (thinker.prompt_expand_func, dit.custom_process_input_func):
|
||||
module_path, _, attr = ref.rpartition(".")
|
||||
module = importlib.import_module(module_path)
|
||||
@@ -1205,7 +1351,8 @@ class TestMingFlashOmniPipeline:
|
||||
assert deploy.connectors is not None
|
||||
assert "shared_memory_connector" in deploy.connectors
|
||||
|
||||
pipeline = _PIPELINE_REGISTRY["ming_flash_omni_image"]
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config("ming_flash_omni_image")
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
stages = merge_pipeline_deploy(pipeline, deploy)
|
||||
assert len(stages) == 2
|
||||
# Stage 0 thinker: AR worker that emits latents.
|
||||
@@ -1356,12 +1503,16 @@ class TestCLIOverrideFlow:
|
||||
"""Test --stage-overrides JSON merge into StageConfig."""
|
||||
|
||||
def test_stage_overrides_merge(self):
|
||||
pipeline = _PIPELINE_REGISTRY["qwen3_omni_moe"]
|
||||
deploy_path = Path(__file__).parent.parent / "vllm_omni" / "deploy" / "qwen3_omni_moe.yaml"
|
||||
if not deploy_path.exists():
|
||||
pytest.skip("Deploy config not found")
|
||||
|
||||
deploy = load_deploy_config(deploy_path)
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
stages = merge_pipeline_deploy(pipeline, deploy)
|
||||
|
||||
# Simulate --stage-overrides '{"0": {"gpu_memory_utilization": 0.5}}'
|
||||
@@ -1370,12 +1521,16 @@ class TestCLIOverrideFlow:
|
||||
assert stages[0].runtime_overrides["gpu_memory_utilization"] == 0.5
|
||||
|
||||
def test_global_override_applies_to_all(self):
|
||||
pipeline = _PIPELINE_REGISTRY["qwen3_omni_moe"]
|
||||
deploy_path = Path(__file__).parent.parent / "vllm_omni" / "deploy" / "qwen3_omni_moe.yaml"
|
||||
if not deploy_path.exists():
|
||||
pytest.skip("Deploy config not found")
|
||||
|
||||
deploy = load_deploy_config(deploy_path)
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
stages = merge_pipeline_deploy(pipeline, deploy)
|
||||
|
||||
overrides = {"enforce_eager": True}
|
||||
@@ -1392,8 +1547,11 @@ class TestAuraOmniDeploy:
|
||||
assert deploy.pipeline == "aura_omni"
|
||||
|
||||
def test_aura_omni_deploy_resolves_four_native_stages(self):
|
||||
pipeline_cfg = StageConfigFactory.resolve_pipeline_config("aura_omni")
|
||||
|
||||
stages = StageConfigFactory._create_from_registry(
|
||||
"qwen3_tts",
|
||||
pipeline_cfg,
|
||||
cli_overrides={},
|
||||
deploy_config_path=str(Path(__file__).parent.parent / "vllm_omni" / "deploy" / "aura_omni.yaml"),
|
||||
)
|
||||
@@ -1417,8 +1575,14 @@ class TestSentinelDefaultPrecedence:
|
||||
to YAML / dataclass defaults (#3035)."""
|
||||
|
||||
def _stages(self, cli_overrides):
|
||||
model_type = "qwen3_omni_moe"
|
||||
pipeline_cfg = StageConfigFactory.resolve_pipeline_config(
|
||||
model_type,
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
return StageConfigFactory._create_from_registry(
|
||||
"qwen3_omni_moe",
|
||||
pipeline_cfg,
|
||||
cli_overrides=cli_overrides,
|
||||
)
|
||||
|
||||
@@ -1483,7 +1647,8 @@ class TestSentinelDefaultPrecedence:
|
||||
"""A single ``qwen3_tts`` pipeline picks per-chunk vs end-to-end
|
||||
processors based on ``deploy.async_chunk``, without needing a
|
||||
separate variant pipeline registration."""
|
||||
pipeline = _PIPELINE_REGISTRY["qwen3_tts"]
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config("qwen3_tts")
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
|
||||
# async_chunk=True → stage 0's per-chunk processor wires up, stage 1
|
||||
# has no sync input processor.
|
||||
@@ -1552,9 +1717,8 @@ class TestSentinelDefaultPrecedence:
|
||||
``thinker2talker_token_only`` placeholder (sync_process_input_func).
|
||||
Merge under either async_chunk mode must not re-introduce a
|
||||
stage-0 full-payload hook."""
|
||||
from vllm_omni.config.stage_config import DeployConfig, merge_pipeline_deploy
|
||||
|
||||
pipeline = _PIPELINE_REGISTRY["ming_flash_omni"]
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config("ming_flash_omni")
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
|
||||
stage0, stage1 = pipeline.stages
|
||||
assert stage0.custom_process_next_stage_input_func is None, (
|
||||
@@ -1587,15 +1751,30 @@ class TestSamplingConstraintsPrecedence:
|
||||
"""Test that pipeline sampling_constraints override deploy defaults."""
|
||||
|
||||
def test_constraints_win(self):
|
||||
pipeline = _PIPELINE_REGISTRY["qwen3_omni_moe"]
|
||||
deploy_path = Path(__file__).parent.parent / "vllm_omni" / "deploy" / "qwen3_omni_moe.yaml"
|
||||
if not deploy_path.exists():
|
||||
pytest.skip("Deploy config not found")
|
||||
|
||||
deploy = load_deploy_config(deploy_path)
|
||||
pipeline = StageConfigFactory.resolve_pipeline_config(
|
||||
"qwen3_omni_moe",
|
||||
Q3_OMNI_ALL_STAGES_HF_CONFIG,
|
||||
)
|
||||
assert isinstance(pipeline, PipelineConfig)
|
||||
stages = merge_pipeline_deploy(pipeline, deploy)
|
||||
|
||||
# Pipeline says detokenize=True for thinker, deploy can't override
|
||||
assert stages[0].yaml_extras["default_sampling_params"]["detokenize"] is True
|
||||
# Pipeline says stop_token_ids=[2150] for talker
|
||||
assert stages[1].yaml_extras["default_sampling_params"]["stop_token_ids"] == [2150]
|
||||
|
||||
|
||||
class TestPipelineConfigResolvers:
|
||||
@pytest.mark.parametrize("resolver", [obj for obj in OMNI_PIPELINES.values() if callable(obj)])
|
||||
def test_all_resolvers_reject_bad_types(self, resolver):
|
||||
"""Ensure that all resolvers registered reject incorrect config types."""
|
||||
|
||||
class NotTheRightHfConfig(PretrainedConfig):
|
||||
pass
|
||||
|
||||
assert resolver(NotTheRightHfConfig()) is None
|
||||
@@ -10,7 +10,7 @@ from collections.abc import Mapping
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from vllm_omni.config.stage_config import StageConfigFactory
|
||||
from vllm_omni.config.config_factory import StageConfigFactory
|
||||
from vllm_omni.diffusion.data import (
|
||||
DiffusionParallelConfig,
|
||||
OmniDiffusionConfig,
|
||||
|
||||
@@ -19,7 +19,7 @@ import torch
|
||||
|
||||
class TestHiggsAudioV3Config:
|
||||
def test_default_config_loads(self):
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.configuration_higgs_audio_v3 import (
|
||||
from vllm_omni.transformers_utils.configs.higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
|
||||
@@ -34,7 +34,7 @@ class TestHiggsAudioV3Config:
|
||||
assert config.tie_modality_embeddings is True
|
||||
|
||||
def test_custom_audio_encoder_config(self):
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.configuration_higgs_audio_v3 import (
|
||||
from vllm_omni.transformers_utils.configs.higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
|
||||
@@ -51,7 +51,7 @@ class TestHiggsAudioV3Config:
|
||||
assert config.tie_modality_embeddings is False
|
||||
|
||||
def test_invalid_num_codebooks_rejected(self):
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.configuration_higgs_audio_v3 import (
|
||||
from vllm_omni.transformers_utils.configs.higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
|
||||
@@ -59,7 +59,7 @@ class TestHiggsAudioV3Config:
|
||||
HiggsAudioV3Config(audio_encoder_config={"num_codebooks": 0, "vocab_size": 1026})
|
||||
|
||||
def test_negative_num_codebooks_rejected(self):
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.configuration_higgs_audio_v3 import (
|
||||
from vllm_omni.transformers_utils.configs.higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
|
||||
@@ -67,7 +67,7 @@ class TestHiggsAudioV3Config:
|
||||
HiggsAudioV3Config(audio_encoder_config={"num_codebooks": -1, "vocab_size": 1026})
|
||||
|
||||
def test_special_token_ids_initially_none(self):
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.configuration_higgs_audio_v3 import (
|
||||
from vllm_omni.transformers_utils.configs.higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
|
||||
@@ -77,7 +77,7 @@ class TestHiggsAudioV3Config:
|
||||
assert config.audio_continuation_id is None
|
||||
|
||||
def test_hidden_size_from_text_config(self):
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.configuration_higgs_audio_v3 import (
|
||||
from vllm_omni.transformers_utils.configs.higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
|
||||
@@ -797,12 +797,12 @@ class TestAudioFeedback:
|
||||
class TestCodecStrictness:
|
||||
def test_bundled_missing_quantizer_key_raises(self):
|
||||
"""Bundled codec load must fail when a quantizer codebook key is missing."""
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.configuration_higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.higgs_audio_v3_code2wav import (
|
||||
HiggsAudioV3Code2Wav,
|
||||
)
|
||||
from vllm_omni.transformers_utils.configs.higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
|
||||
config = HiggsAudioV3Config()
|
||||
c2w = HiggsAudioV3Code2Wav(config=config)
|
||||
@@ -821,12 +821,12 @@ class TestCodecStrictness:
|
||||
|
||||
def test_bundled_missing_fc2_raises(self):
|
||||
"""Bundled codec load must fail when fc2 keys are missing."""
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.configuration_higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.higgs_audio_v3_code2wav import (
|
||||
HiggsAudioV3Code2Wav,
|
||||
)
|
||||
from vllm_omni.transformers_utils.configs.higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
|
||||
config = HiggsAudioV3Config()
|
||||
c2w = HiggsAudioV3Code2Wav(config=config)
|
||||
@@ -844,12 +844,12 @@ class TestCodecStrictness:
|
||||
def test_bundled_unknown_decoder_key_rejected(self, monkeypatch):
|
||||
"""Unknown decoder-side key should be rejected as unconsumed."""
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3 import higgs_audio_v3_code2wav as c2w_mod
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.configuration_higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.higgs_audio_v3_code2wav import (
|
||||
HiggsAudioV3Code2Wav,
|
||||
)
|
||||
from vllm_omni.transformers_utils.configs.higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
|
||||
class FakeDAC(torch.nn.Module):
|
||||
def __init__(self):
|
||||
@@ -871,12 +871,12 @@ class TestCodecStrictness:
|
||||
def test_bundled_encoder_side_keys_accepted(self, monkeypatch):
|
||||
"""Known encoder-side keys should not trigger unconsumed-key rejection."""
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3 import higgs_audio_v3_code2wav as c2w_mod
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.configuration_higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
from vllm_omni.model_executor.models.higgs_audio_v3.higgs_audio_v3_code2wav import (
|
||||
HiggsAudioV3Code2Wav,
|
||||
)
|
||||
from vllm_omni.transformers_utils.configs.higgs_audio_v3 import (
|
||||
HiggsAudioV3Config,
|
||||
)
|
||||
|
||||
class FakeDAC(torch.nn.Module):
|
||||
def __init__(self):
|
||||
@@ -1199,9 +1199,9 @@ class TestRegistry:
|
||||
assert "HiggsAudioV3Code2WavForConditionalGeneration" in _OMNI_MODELS
|
||||
|
||||
def test_pipeline_registered(self):
|
||||
from vllm_omni.config.pipeline_registry import _OMNI_PIPELINES
|
||||
from vllm_omni.config.pipeline_registry import OMNI_PIPELINES
|
||||
|
||||
assert "higgs_multimodal_qwen3" in _OMNI_PIPELINES
|
||||
assert "higgs_multimodal_qwen3" in OMNI_PIPELINES
|
||||
|
||||
def test_deploy_yaml_exists(self):
|
||||
import os
|
||||
|
||||
@@ -8,11 +8,11 @@ import pytest
|
||||
import yaml
|
||||
from vllm.utils.argparse_utils import FlexibleArgumentParser
|
||||
|
||||
from vllm_omni.config.config_factory import StageConfigFactory
|
||||
from vllm_omni.config.pipeline_registry import OMNI_PIPELINES
|
||||
from vllm_omni.config.stage_config import (
|
||||
_PIPELINE_REGISTRY,
|
||||
DeployConfig,
|
||||
PipelineConfig,
|
||||
StageConfigFactory,
|
||||
StageDeployConfig,
|
||||
StagePipelineConfig,
|
||||
)
|
||||
@@ -44,9 +44,9 @@ _TEST_DEPLOY = DeployConfig(
|
||||
@pytest.fixture()
|
||||
def mock_stages(monkeypatch):
|
||||
"""Register a fake pipeline and mock deploy YAML loading."""
|
||||
monkeypatch.setitem(_PIPELINE_REGISTRY._loaded, _TEST_MODEL, _TEST_PIPELINE)
|
||||
monkeypatch.setitem(OMNI_PIPELINES, _TEST_MODEL, _TEST_PIPELINE)
|
||||
monkeypatch.setattr(
|
||||
"vllm_omni.config.stage_config.load_deploy_config",
|
||||
"vllm_omni.config.config_factory.load_deploy_config",
|
||||
lambda _path: _TEST_DEPLOY,
|
||||
)
|
||||
return __file__
|
||||
@@ -462,6 +462,7 @@ def test_explicit_cli_arg_reaches_runtime_overrides(mock_stages):
|
||||
explicit_kwargs = ns.get_explicit_kwargs_dict()
|
||||
stages = StageConfigFactory._create_from_registry(
|
||||
_TEST_MODEL,
|
||||
_TEST_PIPELINE,
|
||||
explicit_kwargs,
|
||||
deploy_config_path=mock_stages,
|
||||
)
|
||||
@@ -478,6 +479,7 @@ def test_omitted_default_not_in_runtime_overrides(mock_stages):
|
||||
explicit_kwargs = ns.get_explicit_kwargs_dict()
|
||||
stages = StageConfigFactory._create_from_registry(
|
||||
_TEST_MODEL,
|
||||
_TEST_PIPELINE,
|
||||
explicit_kwargs,
|
||||
deploy_config_path=mock_stages,
|
||||
)
|
||||
@@ -499,6 +501,7 @@ def test_config_file_args_reach_runtime_overrides(mock_stages):
|
||||
explicit_kwargs = ns.get_explicit_kwargs_dict()
|
||||
stages = StageConfigFactory._create_from_registry(
|
||||
_TEST_MODEL,
|
||||
_TEST_PIPELINE,
|
||||
explicit_kwargs,
|
||||
deploy_config_path=mock_stages,
|
||||
)
|
||||
@@ -516,6 +519,7 @@ def test_per_stage_override_routes_correctly(mock_stages):
|
||||
explicit_kwargs = ns.get_explicit_kwargs_dict()
|
||||
stages = StageConfigFactory._create_from_registry(
|
||||
_TEST_MODEL,
|
||||
_TEST_PIPELINE,
|
||||
explicit_kwargs,
|
||||
deploy_config_path=mock_stages,
|
||||
)
|
||||
@@ -537,6 +541,7 @@ def test_explicit_args_omitted_from_yaml(mock_stages):
|
||||
explicit_kwargs = ns.get_explicit_kwargs_dict()
|
||||
stages = StageConfigFactory._create_from_registry(
|
||||
_TEST_MODEL,
|
||||
_TEST_PIPELINE,
|
||||
explicit_kwargs,
|
||||
deploy_config_path=mock_stages,
|
||||
)
|
||||
|
||||
Reference in new issue
Block a user