vllm-omni/pyproject.toml

[build-system]
requires = [
    "setuptools>=77.0.3,<81.0.0",
    "wheel",
    "setuptools-scm>=8.0",
]
build-backend = "setuptools.build_meta"

[project]
name = "vllm-omni"
dynamic = ["version", "dependencies"]
description = "A framework for efficient model inference with omni-modality models"
readme = "README.md"
requires-python = ">=3.10,<3.14"
license = "Apache-2.0"
authors = [
    {name = "vLLM-Omni Team"}
]
keywords = ["vllm", "multimodal", "diffusion", "transformer", "inference", "serving"]
classifiers = [
    "Development Status :: 3 - Alpha",
    "Intended Audience :: Developers",
    "Intended Audience :: Science/Research",
    "Programming Language :: Python :: 3.10",
    "Programming Language :: Python :: 3.11",
    "Programming Language :: Python :: 3.12",
    "Programming Language :: Python :: 3.13",
    "Topic :: Scientific/Engineering :: Artificial Intelligence",
    "Topic :: Software Development :: Libraries :: Python Modules",
]

# Dependencies are now managed dynamically via setup.py based on detected hardware platform.
# This allows automatic installation of the correct platform-specific dependencies (CUDA/ROCm/CPU/XPU/NPU)
# without requiring extras like [cuda]. See requirements/ directory for platform-specific dependencies.

[project.optional-dependencies]

dev = [
    "pytest==9.1.1",
    "pytest-asyncio==1.4.0",
    "pytest-cov==7.1.0",
    "pytest-mock==3.15.1",
    "websockets==16.1",
    "pytest-subtests==0.15.0",
    "datasets==3.6.0",
    "mypy==1.11.1",
    "pre-commit==4.0.1",
    "openai-whisper==20250625",
    "psutil==7.2.2",
    "soundfile==0.14.0",
    "imageio[ffmpeg]==2.37.4",
    "opencv-python==5.0.0.93",
    "mooncake-transfer-engine-cuda13==0.3.11.post1",
    "av==18.0.0", # for ComfyUI tests
    "openpyxl==3.1.5", # for nightly CI
    "pyttsx3==2.99",
    "opencc==1.4.1",
    "mistune==3.3.4", # for example tests
    "torchmetrics==1.9.0", # for accuracy similarity metrics
    "ray[default]==2.56.1", # for RL test
    "jiwer==4.0.0",
    "zhon==2.1.1",
    "zhconv==1.4.3",
    "scipy==1.18.0",
    "funasr==1.3.29",
    "lpips==0.1.4", # for quantization fp8
    "orjson==3.11.9", # to make v1/embeddings API fast
    "voxcpm==2.0.3", # for VoxCPM2 path-locating in tests
    "torchdiffeq==0.2.5", # for covo_audio Test
    "FlagEmbedding==1.4.0", # for hunyuan_image3 tests
    "kernels==0.14.1", # for accuracy tests comparison with diffusers output (0.15.0 incompatible with diffusers 0.38.0)
    "s3tokenizer==0.3.0", # for tts test
    "step-audio2==1.0.0", # Step-Audio2 / MiniCPM-o 4.5 Token2Wav (flashcosyvoice / hyperpyyaml)
]

demo = [
    "gradio>=6.7.0",
    "opencv-python>=4.12.0.88",  # cli/stream_client frame reading
    "requests>=2.28.0",  # cli/stream_client HTTP
    "cn2an>=0.5.22",  # Chinese number normalization for TTS (e.g. "1次" -> "一次")
]

forced-aligner = [
    "qwen-asr>=0.0.6",  # faithful Japanese/Korean word segmentation for the TTS forced aligner
]

indextts2 = [
    "fugashi>=1.2.0",
    "unidic-lite>=1.0.0",
    "wetext>=0.0.9; sys_platform != 'linux'",
    "WeTextProcessing; sys_platform == 'linux'",
]

# Local-test-only extras
local = [
    "FlagEmbedding",
]

# LongCat-Video-Avatar A2V/AI2V runtime dependencies for vocal
# separation and loudness normalization.
longcat-video-avatar = [
    "audio-separator==0.44.2",
    "pyloudnorm==0.1.1",
    "soundfile>=0.13.1",
]

# Optional Blackwell-only fused-bias FP8 GEMM; package is `quack-kernels` (imported
# as `quack`), not the unrelated `quack`. See docs/user_guide/quantization/fp8.md.
quack = [
    "quack-kernels>=0.3.11",
]

# Optional Blackwell-native FlashAttention-4 backend. Keep this opt-in because
# the CuTe/CUTLASS dependency is CUDA 13 and Blackwell specific.
fa4 = [
    "flash-attn-4[cu13]==4.0.0b18",
]

docs = [
    "mkdocs>=1.5.0",
    "mkdocs-api-autonav",
    "mkdocs-material",
    "mkdocstrings-python",
    "mkdocs-gen-files",
    "mkdocs-awesome-nav",
    "mkdocs-redirects",
    "mkdocs-glightbox",
    "mkdocs-git-revision-date-localized-plugin",
    "mkdocs-minify-plugin",
    "regex",
    "ruff",
    "pydantic",
]

soulx-svs = [
    "funasr>=1.0.0",
    "g2pM>=0.1.2.5",
    "g2p-en>=2.1.0",
    "ToJyutping>=3.2.0",
]

[project.urls]
Homepage = "https://github.com/vllm-project/vllm-omni"
Repository = "https://github.com/vllm-project/vllm-omni"
Documentation = "https://vllm-omni.readthedocs.io"
"Bug Tracker" = "https://github.com/vllm-project/vllm-omni/issues"

[project.scripts]
vllm-omni = "vllm_omni.entrypoints.cli.main:main"

# Register vllm-omni's model registrations on every vllm subprocess. vllm
# auto-loads anything advertised under `vllm.general_plugins` in its own
# `load_general_plugins()`, so worker subprocesses spawned by vllm (which
# only import vllm, not vllm_omni) will still pick up the omni archs.
[project.entry-points."vllm.general_plugins"]
vllm_omni_register_models = "vllm_omni.engine.arg_utils:register_omni_models_to_vllm"


[tool.setuptools.packages.find]
where = ["."]
include = ["vllm_omni*"]

[tool.setuptools.package-data]
"vllm_omni" = ["_version.py", "py.typed"]

[tool.setuptools_scm]
# no extra settings needed, presence enables setuptools-scm

[tool.ruff]
line-length = 120
exclude = [
    ".eggs",
    ".git",
    ".hg",
    ".mypy_cache",
    ".tox",
    ".venv",
    "build",
    "dist",
    "vllm_omni.egg-info",
]

[tool.ruff.lint]
select = [
    "E",      # pycodestyle errors
    "W",      # pycodestyle warnings
    "F",      # pyflakes
    "I",      # isort (handled separately, but included for compatibility)
    "N",      # pep8-naming
    "UP",     # pyupgrade
    "TID251", # flake8-tidy-imports.banned-api
]
ignore = [
    "E203",  # whitespace before ':' (conflicts with black)
    # W503 is not needed in ruff as it's compatible with black by default
    "N801",  # class names should use CapWords convention
    "N802",  # function name should follow snake_case
    "N806",  # variable in function should follow snake_case
    "N812",  # lowercase imported as non-lowercase: functional as F
]

[tool.ruff.lint.per-file-ignores]
"examples/**" = ["E501"]  # Allow long lines in examples
"tests/**" = ["E501"]  # Allow long lines in tests
# MiniCPM-o reference modeling code vendored from upstream HF/transformers/MiniCPM-o.
# These files keep upstream conventions (star imports, long lines, math-symbol
# argument names) so they stay easy to diff against upstream releases.
"vllm_omni/model_executor/models/minicpmo_4_5/minicpmo_4_5_omni_llm.py" = ["E402", "E501", "F403", "F405", "N803"]
# NemotronVoiceChat inference modules vendored from NVIDIA-NeMo/Speech (branch
# nemotron-labs-voicechat). Numerics-bearing code is kept byte-close to
# upstream to stay diffable and parity-safe, so upstream style (long lines,
# legacy typing, whitespace, unused-import shims) is exempted. TID251 and the
# F-correctness rules still apply.
"vllm_omni/model_executor/models/nemotron_voicechat/nemo_vendored/**" = ["E501", "E721", "E741", "F401", "F841", "I001", "N803", "N814", "UP006", "UP007", "UP008", "UP011", "UP032", "UP035", "W291", "W293"]
"vllm_omni/experimental/fullduplex/joyvl/decision/prompts.py" = ["E501"]  # System/summarizer prompts are long prose lines

[tool.ruff.lint.flake8-tidy-imports.banned-api]
"librosa".msg = "The librosa module is banned, use vllm.multimodal helpers instead"
"torch.cuda.max_memory_allocated".msg = "Use torch.accelerator.max_memory_allocated"
"torch.cuda.reset_peak_memory_stats".msg = "Use torch.accelerator.reset_peak_memory_stats"
"torch.cuda.device_count".msg = "Use torch.accelerator.device_count"
"torch.cuda.current_device".msg = "Use torch.accelerator.current_device_index"
"torch.cuda.empty_cache".msg  = "Use torch.accelerator.empty_cache"
"torch.cuda.synchronize".msg = "Use torch.accelerator.synchronize"

[tool.mypy]
python_version = "3.12, 3.13"
warn_return_any = true
warn_unused_configs = true
disallow_untyped_defs = true
disallow_incomplete_defs = true
check_untyped_defs = true
disallow_untyped_decorators = true
no_implicit_optional = true
warn_redundant_casts = true
warn_unused_ignores = true
warn_no_return = true
warn_unreachable = true
strict_equality = true

[tool.pytest.ini_options]
asyncio_mode = "auto"
testpaths = ["tests"]
python_files = ["test_*.py", "*_test.py"]
python_classes = ["Test*"]
python_functions = ["test_*"]
addopts = [
    "--strict-markers",
    "--strict-config"
]
markers = [
    # ci/cd required
    "core_model: L1&L2 tests (run in each PR)",
    "advanced_model: L3 level tests (run on each merge)",
    "full_model: L4 level tests (run nightly)",
    "local_model: tests (run locally)",
    # function module markers
    "diffusion: Diffusion model tests",
    "omni: Omni model tests",
    "tts: TTS model tests",
    "cache: Cache backend tests",
    "parallel: Parallelism/distributed tests",
    "sp: Sequence parallelism tests",
    "example: Doc example code tests",
    # platform markers
    "cpu: Tests that run on CPU",
    "gpu: Tests that run on GPU (auto-added)",
    "cuda: Tests that run on CUDA (auto-added)",
    "rocm: Tests that run on AMD/ROCm (auto-added)",
    "xpu: Tests that run on XPU (auto-added)",
    "npu: Tests that run on NPU/Ascend (auto-added)",
    "musa: Tests that run on MUSA/Moore Threads (auto-added)",
    # specified computation resources marks (auto-added)
    # Tag with [hardware-resource] for tests.helpers.mark.get_hardware_mark_list().
    "H100: [hardware-resource] Tests that require H100 GPU",
    "H200: [hardware-resource] Tests that require H200 GPU",
    "L4: [hardware-resource] Tests that require L4 GPU",
    "B200: [hardware-resource] Tests that require B200 GPU",
    "MI325: [hardware-resource] Tests that require MI325 GPU (AMD/ROCm)",
    "B60: [hardware-resource] Tests that require Intel Arc Pro B60 XPU",
    "S5000: [hardware-resource] Tests that require S5000 GPU (Moore Threads/MUSA)",
    "A2: [hardware-resource] Tests that require A2 NPU",
    "A3: [hardware-resource] Tests that require A3 NPU",
    "distributed_cuda: Tests that require multi cards on CUDA platform",
    "distributed_rocm: Tests that require multi cards on ROCm platform",
    "distributed_xpu: Tests that require multi cards on XPU platform",
    "distributed_npu: Tests that require multi cards on NPU platform",
    "distributed_musa: Tests that require multi cards on MUSA platform",
    "skipif_cuda: Skip if the num of CUDA cards is less than the required",
    "skipif_rocm: Skip if the num of ROCm cards is less than the required",
    "skipif_xpu: Skip if the num of XPU cards is less than the required",
    "skipif_npu: Skip if the num of NPU cards is less than the required",
    "skipif_musa: Skip if the num of MUSA cards is less than the required",
    # more detailed markers
    "sp: Sequence parallelism tests (multi-GPU)",
    "slow: Slow tests (may skip in quick CI)",
    "benchmark: Benchmark tests",
]
filterwarnings = [
    "ignore:.*does not have '__test__' attribute.*:UserWarning",
    "ignore:.*does not have '__bases__' attribute.*:UserWarning",
]

[tool.coverage.run]
core = "sysmon"
source = ["vllm_omni"]
# Online-serving tests launch the server via subprocess.Popen and stop it with
# SIGTERM. pytest-cov no longer instruments subprocesses implicitly, and a
# SIGTERM-ed process discards its data, so without patch/sigterm a coverage run
# over an online test measures only the pytest parent.
patch = ["subprocess"]
concurrency = ["multiprocessing", "thread"]
sigterm = true

[tool.coverage.report]
show_missing = true
skip_covered = true

[tool.typos.default]
extend-ignore-identifiers-re = [
    ".*_thw",
    ".*thw",
    "ein",
    ".*arange",
    ".*MoBA",
    ".*temperal_downsample",
    ".*_nd",
    ".*nd$",
    ".*EmbedND$",
    ".*nothink.*",
    ".*NOTHINK.*",
    ".*nin.*",
    ".*[Oo]no_[Aa]nna.*",
    ".*cann.*",
    ".*[Nn][Dd]3.*",
    ".*updator.*",
    "byt5.*",
    # NemotronVoiceChat vendored triton kernel: dout/DOUT abbreviate d_out
    # (output feature dim), not "doubt".
    ".*dout.*",
    ".*DOUT.*",
]
extend-ignore-re = [
    "\\.\\.\\.n,d->\\.\\.\\.nd",
]
[tool.typos.default.extend-words]
ue = "ue"
semantics = "semantics"
fullset = "fullset"
Vai = "Vai"
tockens = "tockens"
CANN = "CANN"
# Step-Audio reasoning parser streams `</think>` tokens split across chunks
# (e.g. "</" + "thi" + "nk>"). The fragment "thi" appears literally in its
# parser logic and tests and is not a typo of "the"/"this".
thi = "thi"
ND = "ND"
byt5 = "byt5"
# "seeked" is the standard HTML media element DOM event name (fires when a
# seek completes), used by the word-timestamps Gradio demo. Not a typo.
seeked = "seeked"
# Ryan Leary: author credited in the NemotronVoiceChat vendored NeMo
# features.py copyright header. A surname, not a typo of "leery".
Leary = "Leary"
# IndexTTS 2.5 language codes and speech-emotion-recognition event token.
fo = "fo"
ba = "ba"
SER = "SER"