Skip to content

vllm_omni.host_weight_runtime

Neutral contracts for immutable runtime-ready host weight artifacts.

Modules:

Name Description
config

Typed policy for the loader-adjacent Host Weight Runtime.

errors

Stable Host Weight Runtime failures and actions.

filesystem

Local filesystem Host Weight Store implementation.

identity

Deterministic semantic identity for runtime-ready host weights.

inspection

Typed, read-only snapshots for local Host Weight Store inspection.

lease

Process-local ownership for immutable mmap-backed host tensors.

manifest

Versioned immutable manifests for runtime-ready host weights.

outcomes

Typed store, resolution, and post-load publication outcomes.

protocols

Provider and consumer protocols for Host Weight Runtime extensions.

runtime

Policy-driven exact resolution over HostWeightStore contracts.

AdaptationIdentity dataclass

fingerprint class-attribute instance-attribute

fingerprint: str | None = None

kind class-attribute instance-attribute

kind: str = 'base'

metadata class-attribute instance-attribute

metadata: CanonicalJson = field(
    default_factory=CanonicalJson.empty
)

from_dict classmethod

from_dict(value: dict[str, object]) -> AdaptationIdentity

to_dict

to_dict() -> dict[str, JsonValue]

ArtifactInspection dataclass

artifact_key instance-attribute

artifact_key: str

artifact_lock_active instance-attribute

artifact_lock_active: bool

backing_fingerprint instance-attribute

backing_fingerprint: str | None

build_lock_active instance-attribute

build_lock_active: bool

failure instance-attribute

failure: HostWeightFailure | None

manifest instance-attribute

manifest: WeightManifest | None

publication_id instance-attribute

publication_id: str | None

size_bytes instance-attribute

size_bytes: int

status instance-attribute

status: StoreStatus

ArtifactInventoryEntry dataclass

artifact_key instance-attribute

artifact_key: str

reason class-attribute instance-attribute

reason: str | None = None

size_bytes instance-attribute

size_bytes: int

state instance-attribute

storage_name instance-attribute

storage_name: str

ArtifactInventoryState

Bases: str, Enum

DENIED class-attribute instance-attribute

DENIED = 'denied'

QUARANTINED class-attribute instance-attribute

QUARANTINED = 'quarantined'

READY class-attribute instance-attribute

READY = 'ready'

TEMPORARY class-attribute instance-attribute

TEMPORARY = 'temporary'

ArtifactWriter

Bases: Protocol

open_tensor_file

open_tensor_file(
    relative_name: str,
    specs: tuple[TensorWriteSpec, ...],
    *,
    ordered: bool = False,
) -> TensorFileWriter

AttemptResult

Bases: str, Enum

ERROR class-attribute instance-attribute

ERROR = 'error'

HIT class-attribute instance-attribute

HIT = 'hit'

MISS class-attribute instance-attribute

MISS = 'miss'

SKIPPED class-attribute instance-attribute

SKIPPED = 'skipped'

SUCCESS class-attribute instance-attribute

SUCCESS = 'success'

TIMEOUT class-attribute instance-attribute

TIMEOUT = 'timeout'

BuildRequest dataclass

identity instance-attribute

CanonicalJson dataclass

Immutable snapshot of deterministic JSON metadata.

encoded instance-attribute

encoded: bytes

empty classmethod

empty() -> CanonicalJson

from_value classmethod

from_value(value: object) -> CanonicalJson

to_value

to_value() -> JsonValue

CapacityPolicy dataclass

eviction class-attribute instance-attribute

eviction: str = 'none'

max_artifact_bytes class-attribute instance-attribute

max_artifact_bytes: int | None = None

max_store_bytes class-attribute instance-attribute

max_store_bytes: int | None = None

min_free_bytes class-attribute instance-attribute

min_free_bytes: int = 0

ComponentIdentity dataclass

metadata class-attribute instance-attribute

metadata: CanonicalJson = field(
    default_factory=CanonicalJson.empty
)

name instance-attribute

name: str

ownership instance-attribute

ownership: str

from_dict classmethod

from_dict(value: dict[str, object]) -> ComponentIdentity

to_dict

to_dict() -> dict[str, JsonValue]

CoordinationScope

Bases: str, Enum

ARTIFACT_BUNDLE class-attribute instance-attribute

ARTIFACT_BUNDLE = 'artifact_bundle'

GROUP_COHORT class-attribute instance-attribute

GROUP_COHORT = 'group_cohort'

SINGLE_PROCESS class-attribute instance-attribute

SINGLE_PROCESS = 'single_process'

DomainInspection dataclass

domain_id instance-attribute

domain_id: str

domain_uuid instance-attribute

domain_uuid: str

filesystem_free_bytes instance-attribute

filesystem_free_bytes: int

filesystem_type instance-attribute

filesystem_type: str

inventory instance-attribute

inventory: tuple[ArtifactInventoryEntry, ...]

max_artifact_bytes instance-attribute

max_artifact_bytes: int | None

max_store_bytes instance-attribute

max_store_bytes: int | None

min_free_bytes instance-attribute

min_free_bytes: int

policy_version instance-attribute

policy_version: int

scope instance-attribute

scope: str

storage_class instance-attribute

storage_class: str

store_bytes instance-attribute

store_bytes: int

FailureCode

Bases: str, Enum

ACTIVE_BUILD_TIMEOUT class-attribute instance-attribute

ACTIVE_BUILD_TIMEOUT = 'active_build_timeout'

ACTIVE_LEASE class-attribute instance-attribute

ACTIVE_LEASE = 'active_lease'

ARTIFACT_DENIED class-attribute instance-attribute

ARTIFACT_DENIED = 'artifact_denied'

ARTIFACT_TOO_LARGE class-attribute instance-attribute

ARTIFACT_TOO_LARGE = 'artifact_too_large'

CANONICAL_SOURCE_FAILED class-attribute instance-attribute

CANONICAL_SOURCE_FAILED = 'canonical_source_failed'

CHECKSUM_MISMATCH class-attribute instance-attribute

CHECKSUM_MISMATCH = 'checksum_mismatch'

COHORT_POLICY_MISMATCH class-attribute instance-attribute

COHORT_POLICY_MISMATCH = 'cohort_policy_mismatch'

DOMAIN_POLICY_MISMATCH class-attribute instance-attribute

DOMAIN_POLICY_MISMATCH = 'domain_policy_mismatch'

DOMAIN_UNAVAILABLE class-attribute instance-attribute

DOMAIN_UNAVAILABLE = 'domain_unavailable'

ENOSPC class-attribute instance-attribute

ENOSPC = 'enospc'

IDENTITY_COLLISION class-attribute instance-attribute

IDENTITY_COLLISION = 'identity_collision'

INSUFFICIENT_CAPACITY class-attribute instance-attribute

INSUFFICIENT_CAPACITY = 'insufficient_capacity'

INVALID_CONFIG class-attribute instance-attribute

INVALID_CONFIG = 'invalid_config'

LOCAL_MISS class-attribute instance-attribute

LOCAL_MISS = 'local_miss'

MANIFEST_CORRUPT class-attribute instance-attribute

MANIFEST_CORRUPT = 'manifest_corrupt'

MANIFEST_INCOMPATIBLE class-attribute instance-attribute

MANIFEST_INCOMPATIBLE = 'manifest_incompatible'

NUMA_PLACEMENT_INVALID class-attribute instance-attribute

NUMA_PLACEMENT_INVALID = 'numa_placement_invalid'

PRODUCER_FAILED class-attribute instance-attribute

PRODUCER_FAILED = 'producer_failed'

PRODUCER_UNAVAILABLE class-attribute instance-attribute

PRODUCER_UNAVAILABLE = 'producer_unavailable'

PRODUCER_UNSUPPORTED class-attribute instance-attribute

PRODUCER_UNSUPPORTED = 'producer_unsupported'

PRODUCTION_TIMEOUT class-attribute instance-attribute

PRODUCTION_TIMEOUT = 'production_timeout'

PUBLICATION_FAILED class-attribute instance-attribute

PUBLICATION_FAILED = 'publication_failed'

QUARANTINE_FAILED class-attribute instance-attribute

QUARANTINE_FAILED = 'quarantine_failed'

REMOTE_ARTIFACT_INVALID class-attribute instance-attribute

REMOTE_ARTIFACT_INVALID = 'remote_artifact_invalid'

REMOTE_AUTH_FAILED class-attribute instance-attribute

REMOTE_AUTH_FAILED = 'remote_auth_failed'

REMOTE_MISS class-attribute instance-attribute

REMOTE_MISS = 'remote_miss'

REMOTE_TIMEOUT class-attribute instance-attribute

REMOTE_TIMEOUT = 'remote_timeout'

REMOTE_UNAVAILABLE class-attribute instance-attribute

REMOTE_UNAVAILABLE = 'remote_unavailable'

RESTORER_UNAVAILABLE class-attribute instance-attribute

RESTORER_UNAVAILABLE = 'restorer_unavailable'

RESTORE_COMMIT_FAILED class-attribute instance-attribute

RESTORE_COMMIT_FAILED = 'restore_commit_failed'

RESTORE_PREFLIGHT_FAILED class-attribute instance-attribute

RESTORE_PREFLIGHT_FAILED = 'restore_preflight_failed'

STORAGE_CLASS_MISMATCH class-attribute instance-attribute

STORAGE_CLASS_MISMATCH = 'storage_class_mismatch'

STORE_LIMIT_EXCEEDED class-attribute instance-attribute

STORE_LIMIT_EXCEEDED = 'store_limit_exceeded'

UNSUPPORTED class-attribute instance-attribute

UNSUPPORTED = 'unsupported'

FileManifestEntry dataclass

file_name instance-attribute

file_name: str

sha256 instance-attribute

sha256: str

size instance-attribute

size: int

from_dict classmethod

from_dict(value: dict[str, object]) -> FileManifestEntry

to_dict

to_dict() -> dict[str, JsonValue]

HostWeightError

Bases: RuntimeError

Internal exception carrying a stable, policy-neutral failure.

failure instance-attribute

failure = failure

HostWeightFailure dataclass

code instance-attribute

details class-attribute instance-attribute

details: CanonicalJson = field(
    default_factory=CanonicalJson.empty
)

message instance-attribute

message: str

retryable instance-attribute

retryable: bool

stage instance-attribute

HostWeightInspector

Bases: Protocol

inspect_artifact

inspect_artifact(
    identity: WeightArtifactIdentity,
    *,
    validation: ValidationLevel,
) -> ArtifactInspection

inspect_domain

inspect_domain() -> DomainInspection

HostWeightLease

Stable tensor views and resources retained until transport teardown.

closed property

closed: bool

identity instance-attribute

identity = identity

manifest instance-attribute

manifest = manifest

mapped_regions instance-attribute

mapped_regions = mapped_regions

provenance instance-attribute

provenance = provenance

tensors instance-attribute

tensors = MappingProxyType(self._tensor_storage)

close

close() -> None

HostWeightLeaseCarrier

Process-local, single-take ownership carrier for a host-weight lease.

The loader creates the carrier after a restore transaction has committed. The runner/backend takes the lease exactly once before starting any asynchronous transport work. If setup fails before that handoff, closing the carrier releases the loader-owned lease.

closed property

closed: bool

taken property

taken: bool

close

close() -> None

Close the lease if ownership has not been transferred.

take

take() -> HostWeightLease

Transfer the lease to the transport owner exactly once.

HostWeightResolution dataclass

lease class-attribute instance-attribute

lease: HostWeightLease | None = None

report instance-attribute

HostWeightRuntime

Resolve or publish one exact representation without owning restoration.

config instance-attribute

config = config

enabled property

enabled: bool

store instance-attribute

store = store

from_config classmethod

from_config(
    config: HostWeightRuntimeConfig,
    *,
    observer: Callable[[ResolutionReport], None]
    | None = None,
    publication_observer: Callable[
        [PostLoadPublicationReport], None
    ]
    | None = None,
) -> HostWeightRuntime

Construct lazily so disabled mode performs no filesystem probe.

publish_after_load

publish_after_load(
    identity: WeightArtifactIdentity | None = None,
    *,
    producer: WeightProducer | None = None,
) -> PostLoadPublicationReport

Synchronously warm a future startup without revising resolution.

A successful store operation returns a validated lease. The runtime closes that lease before returning because POST_LOAD_ONLY production must not mutate or rebind the model serving the current startup.

resolve

resolve(
    identity: WeightArtifactIdentity | None = None,
    *,
    producer: WeightProducer | None = None,
) -> HostWeightResolution

HostWeightRuntimeConfig dataclass

capacity class-attribute instance-attribute

capacity: CapacityPolicy = field(
    default_factory=CapacityPolicy
)

domain class-attribute instance-attribute

domain: StorageDomainPolicy | None = None

integrity class-attribute instance-attribute

integrity: IntegrityPolicy = field(
    default_factory=IntegrityPolicy
)

mode class-attribute instance-attribute

production class-attribute instance-attribute

production: ProductionPolicy = field(
    default_factory=ProductionPolicy
)

remote class-attribute instance-attribute

remote: RemoteImportPolicy = field(
    default_factory=RemoteImportPolicy
)

schema_version class-attribute instance-attribute

schema_version: int = 1

wait class-attribute instance-attribute

wait: WaitPolicy = field(default_factory=WaitPolicy)

HostWeightStore

Bases: Protocol

Exact artifact access whose deadlines bound lock acquisition only.

get_or_build

get_or_build(
    request: BuildRequest,
    producer: WeightProducer,
    *,
    validation: ValidationLevel,
    deadline: float,
) -> StoreResult

lookup

lookup(
    identity: WeightArtifactIdentity,
    *,
    validation: ValidationLevel,
    deadline: float | None = None,
) -> StoreResult

IdentityValidationError

Bases: ValueError

Raised when an artifact identity is not deterministic JSON.

IntegrityPolicy dataclass

background_scrub class-attribute instance-attribute

background_scrub: bool = False

local_lookup class-attribute instance-attribute

require_trusted_permissions class-attribute instance-attribute

require_trusted_permissions: bool = True

LeaseProvenance dataclass

artifact_content_sha256 instance-attribute

artifact_content_sha256: str

artifact_origin instance-attribute

artifact_origin: str

backing_fingerprint instance-attribute

backing_fingerprint: str

domain_uuid instance-attribute

domain_uuid: str

identity_digest instance-attribute

identity_digest: str

producer_id instance-attribute

producer_id: str | None

provider_id instance-attribute

provider_id: str | None

publication_id instance-attribute

publication_id: str

resolution_id instance-attribute

resolution_id: str

resolution_source instance-attribute

resolution_source: str

validation_level instance-attribute

validation_level: str

LookupPhase

Bases: str, Enum

POST_LOAD_ONLY class-attribute instance-attribute

POST_LOAD_ONLY = 'post_load_only'

PRE_LOAD_SAFE class-attribute instance-attribute

PRE_LOAD_SAFE = 'pre_load_safe'

ManifestValidationError

Bases: ValueError

Raised when a published manifest violates its schema.

MappedHostRegion dataclass

One process-local virtual range belonging to an artifact payload file.

address instance-attribute

address: int

file_name instance-attribute

file_name: str

length instance-attribute

length: int

read_only class-attribute instance-attribute

read_only: bool = True

PostLoadPublicationOutcome

Bases: str, Enum

Terminal outcome for one explicit post-load publication attempt.

ALREADY_PRESENT class-attribute instance-attribute

ALREADY_PRESENT = 'already_present'

FAILED class-attribute instance-attribute

FAILED = 'failed'

JOINED class-attribute instance-attribute

JOINED = 'joined'

POLICY_DISABLED class-attribute instance-attribute

POLICY_DISABLED = 'policy_disabled'

PUBLISHED class-attribute instance-attribute

PUBLISHED = 'published'

RUNTIME_DISABLED class-attribute instance-attribute

RUNTIME_DISABLED = 'runtime_disabled'

PostLoadPublicationReport dataclass

elapsed_seconds instance-attribute

elapsed_seconds: float

failure class-attribute instance-attribute

failure: HostWeightFailure | None = None

identity_digest instance-attribute

identity_digest: str | None

operation_id instance-attribute

operation_id: str

outcome instance-attribute

ProducerIdentity dataclass

implementation_fingerprint instance-attribute

implementation_fingerprint: str

manifest_schema instance-attribute

manifest_schema: str

producer_id instance-attribute

producer_id: str

restorer_schema instance-attribute

restorer_schema: str

version instance-attribute

version: str

from_dict classmethod

from_dict(value: dict[str, object]) -> ProducerIdentity

to_dict

to_dict() -> dict[str, JsonValue]

ProductionMetadata dataclass

format_metadata class-attribute instance-attribute

format_metadata: CanonicalJson = field(
    default_factory=CanonicalJson.empty
)

producer_schema instance-attribute

producer_schema: str

restorer_schema instance-attribute

restorer_schema: str

ProductionPolicy dataclass

allow_local_build class-attribute instance-attribute

allow_local_build: bool = True

allow_post_load_publish class-attribute instance-attribute

allow_post_load_publish: bool = False

ProductionSourceMode

Bases: str, Enum

CHECKPOINT_STREAM class-attribute instance-attribute

CHECKPOINT_STREAM = 'checkpoint_stream'

FINALIZED_MODEL class-attribute instance-attribute

FINALIZED_MODEL = 'finalized_model'

PublicationMarker dataclass

artifact_content_sha256 instance-attribute

artifact_content_sha256: str

canonical_bytes property

canonical_bytes: bytes

integrity_mode instance-attribute

integrity_mode: str

manifest_sha256 instance-attribute

manifest_sha256: str

publication_id instance-attribute

publication_id: str

schema_version instance-attribute

schema_version: int

from_dict classmethod

from_dict(value: dict[str, object]) -> PublicationMarker

to_dict

to_dict() -> dict[str, JsonValue]

RemoteArtifactDescriptor dataclass

content_sha256 instance-attribute

content_sha256: str

identity instance-attribute

manifest_sha256 instance-attribute

manifest_sha256: str

provider_id instance-attribute

provider_id: str

RemoteArtifactProvider

Bases: Protocol

provider_id property

provider_id: str

fetch

probe

probe(
    identity: WeightArtifactIdentity,
) -> RemoteArtifactDescriptor | None

RemoteImportPolicy dataclass

on_local_miss class-attribute instance-attribute

providers class-attribute instance-attribute

providers: tuple[str, ...] = ()

RemoteOnMiss

Bases: str, Enum

DISABLED class-attribute instance-attribute

DISABLED = 'disabled'

REQUIRE class-attribute instance-attribute

REQUIRE = 'require'

TRY class-attribute instance-attribute

TRY = 'try'

ResolutionAction

Bases: str, Enum

CANONICAL_FALLBACK class-attribute instance-attribute

CANONICAL_FALLBACK = 'canonical_fallback'

FAIL_STARTUP class-attribute instance-attribute

FAIL_STARTUP = 'fail_startup'

QUARANTINE_THEN_PRODUCE class-attribute instance-attribute

QUARANTINE_THEN_PRODUCE = 'quarantine_then_produce'

RETURN_LEASE class-attribute instance-attribute

RETURN_LEASE = 'return_lease'

SKIP class-attribute instance-attribute

SKIP = 'skip'

TRY_PRODUCER class-attribute instance-attribute

TRY_PRODUCER = 'try_producer'

TRY_REMOTE class-attribute instance-attribute

TRY_REMOTE = 'try_remote'

ResolutionAttempt dataclass

action instance-attribute

elapsed_seconds instance-attribute

elapsed_seconds: float

failure class-attribute instance-attribute

failure: HostWeightFailure | None = None

result instance-attribute

result: AttemptResult

stage instance-attribute

ResolutionOutcome

Bases: str, Enum

CANONICAL_DIRECT class-attribute instance-attribute

CANONICAL_DIRECT = 'canonical_direct'

CANONICAL_FALLBACK class-attribute instance-attribute

CANONICAL_FALLBACK = 'canonical_fallback'

FAILED class-attribute instance-attribute

FAILED = 'failed'

LOCAL_HIT class-attribute instance-attribute

LOCAL_HIT = 'local_hit'

LOCAL_PRODUCTION class-attribute instance-attribute

LOCAL_PRODUCTION = 'local_production'

REMOTE_IMPORT class-attribute instance-attribute

REMOTE_IMPORT = 'remote_import'

ResolutionReport dataclass

attempts instance-attribute

attempts: tuple[ResolutionAttempt, ...]

elapsed_seconds instance-attribute

elapsed_seconds: float

outcome instance-attribute

resolution_id instance-attribute

resolution_id: str

ResolutionStage

Bases: str, Enum

CANONICAL_LOADING class-attribute instance-attribute

CANONICAL_LOADING = 'canonical_loading'

CAPACITY class-attribute instance-attribute

CAPACITY = 'capacity'

CONFIGURATION class-attribute instance-attribute

CONFIGURATION = 'configuration'

DOMAIN class-attribute instance-attribute

DOMAIN = 'domain'

LIFECYCLE class-attribute instance-attribute

LIFECYCLE = 'lifecycle'

LOOKUP class-attribute instance-attribute

LOOKUP = 'lookup'

PRODUCTION class-attribute instance-attribute

PRODUCTION = 'production'

REMOTE class-attribute instance-attribute

REMOTE = 'remote'

RESTORATION class-attribute instance-attribute

RESTORATION = 'restoration'

VALIDATION class-attribute instance-attribute

VALIDATION = 'validation'

RuntimeMode

Bases: str, Enum

DISABLED class-attribute instance-attribute

DISABLED = 'disabled'

PREFERRED class-attribute instance-attribute

PREFERRED = 'preferred'

REQUIRED class-attribute instance-attribute

REQUIRED = 'required'

RuntimeWeightLayout dataclass

metadata class-attribute instance-attribute

metadata: CanonicalJson = field(
    default_factory=CanonicalJson.empty
)

name instance-attribute

name: str

sequence_parallel_backend class-attribute instance-attribute

sequence_parallel_backend: str = 'none'

sequence_parallel_size class-attribute instance-attribute

sequence_parallel_size: int = 1

tensor_parallel_rank class-attribute instance-attribute

tensor_parallel_rank: int = 0

tensor_parallel_size class-attribute instance-attribute

tensor_parallel_size: int = 1

from_dict classmethod

from_dict(value: dict[str, object]) -> RuntimeWeightLayout

to_dict

to_dict() -> dict[str, JsonValue]

StorageClass

Bases: str, Enum

DISK class-attribute instance-attribute

DISK = 'disk'

TMPFS class-attribute instance-attribute

TMPFS = 'tmpfs'

StorageDomainPolicy dataclass

domain_id class-attribute instance-attribute

domain_id: str = 'node'

numa_prefault class-attribute instance-attribute

numa_prefault: bool = False

root instance-attribute

root: Path

scope class-attribute instance-attribute

storage_class class-attribute instance-attribute

storage_class: StorageClass = StorageClass.DISK

verify_numa_residency class-attribute instance-attribute

verify_numa_residency: bool = False

StorageScope

Bases: str, Enum

NODE class-attribute instance-attribute

NODE = 'node'

NUMA class-attribute instance-attribute

NUMA = 'numa'

StoreResult dataclass

failure class-attribute instance-attribute

failure: HostWeightFailure | None = None

lease class-attribute instance-attribute

lease: HostWeightLease | None = None

status instance-attribute

status: StoreStatus

StoreStatus

Bases: str, Enum

BUILT class-attribute instance-attribute

BUILT = 'built'

FAILED class-attribute instance-attribute

FAILED = 'failed'

HIT class-attribute instance-attribute

HIT = 'hit'

INVALID class-attribute instance-attribute

INVALID = 'invalid'

JOINED class-attribute instance-attribute

JOINED = 'joined'

MISS class-attribute instance-attribute

MISS = 'miss'

TIMEOUT class-attribute instance-attribute

TIMEOUT = 'timeout'

TensorFileWriter

Bases: Protocol

close

close() -> None

write_rows

write_rows(
    name: str, start_row: int, tensor: Tensor
) -> None

write_tensor

write_tensor(name: str, tensor: Tensor) -> None

TensorKind

Bases: str, Enum

BUFFER class-attribute instance-attribute

BUFFER = 'buffer'

PARAMETER class-attribute instance-attribute

PARAMETER = 'parameter'

TensorManifestEntry dataclass

dtype instance-attribute

dtype: str

file_name instance-attribute

file_name: str

kind instance-attribute

kind: TensorKind

name instance-attribute

name: str

nbytes instance-attribute

nbytes: int

role instance-attribute

role: str

sha256 instance-attribute

sha256: str

shape instance-attribute

shape: tuple[int, ...]

storage_key instance-attribute

storage_key: str

stride instance-attribute

stride: tuple[int, ...]

from_dict classmethod

from_dict(value: dict[str, object]) -> TensorManifestEntry

semantic_dict

semantic_dict() -> dict[str, JsonValue]

Return content identity independent of physical file sharding.

to_dict

to_dict() -> dict[str, JsonValue]

TensorWriteSpec dataclass

dtype instance-attribute

dtype: dtype

kind class-attribute instance-attribute

name instance-attribute

name: str

nbytes property

nbytes: int

numel property

numel: int

role class-attribute instance-attribute

role: str = 'weight'

shape instance-attribute

shape: tuple[int, ...]

ValidationLevel

Bases: str, Enum

FS_VERITY class-attribute instance-attribute

FS_VERITY = 'fs_verity'

FULL_CHECKSUM class-attribute instance-attribute

FULL_CHECKSUM = 'full_checksum'

MANIFEST_AND_METADATA class-attribute instance-attribute

MANIFEST_AND_METADATA = 'manifest_and_metadata'

WaitPolicy dataclass

coordination_timeout_seconds class-attribute instance-attribute

coordination_timeout_seconds: float = 600.0

WeightArtifactIdentity dataclass

adaptation instance-attribute

adaptation: AdaptationIdentity

canonical_bytes property

canonical_bytes: bytes

component instance-attribute

component: ComponentIdentity

key property

key: str

layout instance-attribute

producer instance-attribute

producer: ProducerIdentity

representation instance-attribute

representation: WeightRepresentation

schema_version instance-attribute

schema_version: int

source instance-attribute

from_dict classmethod

from_dict(
    value: dict[str, object],
) -> WeightArtifactIdentity

to_dict

to_dict() -> dict[str, JsonValue]

WeightManifest dataclass

artifact_content_sha256 instance-attribute

artifact_content_sha256: str

artifact_key instance-attribute

artifact_key: str

canonical_bytes property

canonical_bytes: bytes

files instance-attribute

files: tuple[FileManifestEntry, ...]

format_metadata class-attribute instance-attribute

format_metadata: CanonicalJson = field(
    default_factory=CanonicalJson.empty
)

identity instance-attribute

producer_schema instance-attribute

producer_schema: str

restorer_schema instance-attribute

restorer_schema: str

schema_version instance-attribute

schema_version: int

tensors instance-attribute

tensors: tuple[TensorManifestEntry, ...]

total_tensor_bytes property

total_tensor_bytes: int

from_dict classmethod

from_dict(value: dict[str, object]) -> WeightManifest

to_dict

to_dict() -> dict[str, JsonValue]

WeightProducer

Bases: Protocol

Synchronous V1 producer; callers do not preempt an active build.

spec property

produce

produce(writer: ArtifactWriter) -> ProductionMetadata

Produce one exact artifact synchronously or raise.

WeightProductionSpec dataclass

coordination_scope class-attribute instance-attribute

lookup_phase class-attribute instance-attribute

outputs instance-attribute

producer_id instance-attribute

producer_id: str

source_mode instance-attribute

source_mode: ProductionSourceMode

WeightRepresentation dataclass

dtype instance-attribute

dtype: str

metadata class-attribute instance-attribute

metadata: CanonicalJson = field(
    default_factory=CanonicalJson.empty
)

name instance-attribute

name: str

from_dict classmethod

from_dict(value: dict[str, object]) -> WeightRepresentation

to_dict

to_dict() -> dict[str, JsonValue]

WeightRestorePlan

Bases: Protocol

A validated, one-shot model mutation plan.

commit

commit() -> None

Mutate the target model exactly once or raise.

WeightRestorer

Bases: Protocol

Validate a lease-to-model restore without mutating either input.

schema property

schema: str

plan_restore

plan_restore(
    model: object, lease: HostWeightLease
) -> WeightRestorePlan

Return a validated plan; all model mutation is deferred to commit().

WeightSourceIdentity dataclass

fingerprint instance-attribute

fingerprint: str

metadata class-attribute instance-attribute

metadata: CanonicalJson = field(
    default_factory=CanonicalJson.empty
)

model_id instance-attribute

model_id: str

revision instance-attribute

revision: str

from_dict classmethod

from_dict(value: dict[str, object]) -> WeightSourceIdentity

to_dict

to_dict() -> dict[str, JsonValue]