vllm_omni.config.omni_config ¶

cache_config: OmniStageCacheConfig = field(
    default_factory=OmniStageCacheConfig
)

cfg_kv_collect_func: str | None

connector_config: OmniStageConnectorConfig = field(
    default_factory=OmniStageConnectorConfig
)

custom_process_input_func: str | None

custom_process_next_stage_input_func: str | None

engine_output_type: str | None

final_output: bool

final_output_type: str | None

hf_config_name: str | None

input_sources: list[int]

is_comprehension: bool

load_config: OmniStageLoadConfig = field(
    default_factory=OmniStageLoadConfig
)

model_config: OmniStageModelConfig = field(
    default_factory=OmniStageModelConfig
)

model_stage: str

parallel_config: OmniStageParallelConfig = field(
    default_factory=OmniStageParallelConfig
)

prompt_expand_func: str | None

quantization_config: _QuantizationConfigType = None

requires_multimodal_data: bool

runtime_config: OmniStageRuntimeConfig = field(
    default_factory=OmniStageRuntimeConfig
)

scheduler_cls: str | None

scheduler_config: OmniStageSchedulerConfig = field(
    default_factory=OmniStageSchedulerConfig
)

stage_id: int

stage_pipeline_config: StagePipelineConfig

stage_type: StageType

worker_type: str | None

disable_hybrid_kv_cache_manager: bool = False

enable_prefix_caching: bool = False

gpu_memory_utilization: float = Field(
    default=0.9, gt=0.0, le=1.0
)

mm_processor_cache_gb: float | None = Field(
    default=None, ge=0.0
)

input_connectors: dict[str, Any] | None = None

output_connectors: dict[str, Any] | None = None

stage_connector: dict[str, Any] = field(
    default_factory=lambda: {
        "name": "SharedMemoryConnector",
        "extra": {},
    }
)

cfg_parallel_size: int = Field(default=1, ge=1, le=3)

hsdp_replicate_size: int = Field(default=1, ge=1)

hsdp_shard_size: int = -1

mask_sp_padding: bool = False

ring_degree: int = Field(default=1, ge=1)

sequence_parallel_size: int = Field(
    default=1, ge=1, init=False
)

ulysses_degree: int = Field(default=1, ge=1)

ulysses_mode: str = 'strict'

use_hsdp: bool = False

vae_patch_parallel_size: int = Field(default=1, ge=1)

config_format: str | None = None

load_format: str = 'auto'

skip_mm_profiling: bool | None = None

tokenizer_mode: str = 'auto'

active_stream_window: int = Field(default=0, ge=0)

codec_frame_rate_hz: float | None = None

compilation_config: dict[str, Any] | None = None

custom_voice_dir: str | None = None

default_sampling_params: dict[str, Any] | None = None

disable_autocast: bool = False

enable_flashinfer_autotune: bool | None = None

enable_multithread_weight_load: bool = True

enable_sleep_mode: bool = False

enforce_eager: bool = False

has_sampling_extra_args: bool = False

num_weight_load_threads: int = Field(default=4, ge=1)

subtalker_sampling_params: dict[str, Any] | None = None

task_type: str | None = None

data_parallel_size: int = Field(default=1, ge=1)

enable_expert_parallel: bool = False

pipeline_parallel_size: int = Field(default=1, ge=1)

tensor_parallel_size: int = Field(default=1, ge=1)

world_size: int = Field(default=1, ge=1, init=False)

devices: str | None = None

env: dict[str, Any] | None = None

log_level: str = 'info'

log_stats: bool = False

num_gpus: int = Field(default=1, ge=1)

num_replicas: int = Field(default=1, ge=1)

profiler_config: dict[str, Any] | None = None

async_scheduling: bool = True

enable_chunked_prefill: bool = False

max_model_len: int | None = Field(default=None, ge=-1)

max_num_batched_tokens: int | None = Field(
    default=None, ge=1
)

max_num_seqs: int = Field(default=128, ge=1)

orchestrator_config: VllmOmniOrchestratorConfig = field(
    default_factory=VllmOmniOrchestratorConfig
)

pipeline_config: PipelineConfig

stage_configs: tuple[StageConfigType, ...]

diffusion_config: _DiffusionConfigProjection = field(
    default_factory=_DiffusionConfigProjection
)

parallel_config: OmniStageDiffusionParallelConfig = field(
    default_factory=OmniStageDiffusionParallelConfig
)

batch_timeout: int = Field(default=10, ge=0)

deploy_config_path: str | None = None

init_timeout: int = Field(default=600, ge=1)

omni_dp_size_local: int = Field(default=1, ge=1)

omni_heartbeat_timeout: float = Field(default=30.0, gt=0.0)

omni_lb_policy: str = 'random'

omni_master_address: str | None = None

omni_master_port: int | None = None

ray_address: str | None = None

shm_threshold_bytes: int = Field(default=65536, ge=0)

stage_init_timeout: int = Field(default=300, ge=1)

worker_backend: str = 'multi_process'