dyff-schema 0.33.0__tar.gz → 0.34.0__tar.gz

{dyff_schema-0.33.0 → dyff_schema-0.34.0}/.gitlab-ci.yml

@@ -8,7 +8,7 @@ stages:
 
 include:
   - project: saferatday0/library/gitlab
-    ref: 0.17.2
+    ref: 0.18.3
     file:
       - gitlab-release.yml
   - project: saferatday0/library/detect-secrets
@@ -20,7 +20,7 @@ include:
     file:
       - prettier.yml
   - project: saferatday0/library/python
-    ref: 0.21.2
+    ref: 0.22.0
     file:
       - python-autoflake.yml
       - python-black.yml

{dyff_schema-0.33.0/dyff_schema.egg-info → dyff_schema-0.34.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dyff-schema
-Version: 0.33.0
+Version: 0.34.0
 Summary: Data models for the Dyff AI auditing platform.
 Author-email: Digital Safety Research Institute <contact@dsri.org>
 License: Apache-2.0

dyff_schema-0.34.0/dyff/schema/_version.py

@@ -0,0 +1,2 @@
+__version__ = version = "0.34.0"
+__version_tuple__ = version_tuple = (0, 34, 0)

{dyff_schema-0.33.0 → dyff_schema-0.34.0}/dyff/schema/v0/r1/oci.py

@@ -181,3 +181,15 @@ class ImageManifest(_OCISchemaBaseModel):
                     " if .artifactType is unspecified"
                 )
         return self
+
+
+__all__ = [
+    "Descriptor",
+    "ExecutionParameters",
+    "HistoryEntry",
+    "ImageConfig",
+    "ImageDescriptor",
+    "ImageManifest",
+    "Platform",
+    "RootFS",
+]

{dyff_schema-0.33.0 → dyff_schema-0.34.0}/dyff/schema/v0/r1/platform.py

@@ -24,6 +24,7 @@ import enum
 import urllib.parse
 from datetime import datetime
 from enum import Enum
+from pathlib import Path
 from typing import Any, Literal, NamedTuple, Optional, Type, Union
 
 import pyarrow
@@ -194,7 +195,8 @@ class Entities(str, enum.Enum):
 
     Account = "Account"
     Analysis = "Analysis"
-    Artifact = "Artifact"
+    # FIXME: (schema v1) Rename to Artifact
+    Artifact = "OCIArtifact"
     Audit = "Audit"
     AuditProcedure = "AuditProcedure"
     Concern = "Concern"
@@ -262,6 +264,9 @@ class Resources(str, enum.Enum):
     @staticmethod
     def for_kind(kind: Entities) -> "Resources":
         try:
+            # FIXME: (schema v1) Special case for legacy OCIArtifact name
+            if kind == Entities.Artifact:
+                return Resources.Artifact
             return Resources[kind.value]
         except KeyError:
             raise ValueError(f"No Resources for Entity kind: {kind}")
@@ -269,7 +274,6 @@ class Resources(str, enum.Enum):
 
 EntityKindLiteral = Literal[
     "Analysis",
-    "Artifact",
     "Audit",
     "AuditProcedure",
     "DataSource",
@@ -284,6 +288,8 @@ EntityKindLiteral = Literal[
     "Method",
     "Model",
     "Module",
+    # FIXME: (schema v1) Rename to Artifact
+    "OCIArtifact",
     "Report",
     "Revision",
     "SafetyCase",
@@ -312,6 +318,9 @@ class EntityIdentifier(DyffSchemaBaseModel):
         description="The .kind of the entity.",
     )
 
+    def resource_path(self) -> str:
+        return f"{Resources.for_kind(Entities(self.kind)).value}/{self.id}"
+
 
 def LabelKey() -> type[str]:
     return Annotated[
@@ -492,7 +501,6 @@ class DyffEntityMetadata(DyffSchemaBaseModel):
 class DyffEntity(Status, Labeled, SchemaVersion, DyffModelWithID):
     kind: Literal[
         "Analysis",
-        "Artifact",
         "Audit",
         "AuditProcedure",
         "DataSource",
@@ -507,6 +515,8 @@ class DyffEntity(Status, Labeled, SchemaVersion, DyffModelWithID):
         "Method",
         "Model",
         "Module",
+        # FIXME: (schema v1) Rename to Artifact
+        "OCIArtifact",
         "Report",
         "Revision",
         "SafetyCase",
@@ -1076,6 +1086,32 @@ class Dataset(DyffEntity, DatasetBase):
         return None
 
 
+# ----------------------------------------------------------------------------
+# Model
+
+
+class File(DyffSchemaBaseModel):
+    mediaType: Optional[str] = pydantic.Field(
+        default=None, description="The media type (MIME type) of the file"
+    )
+    path: str = pydantic.Field(
+        description="The relative path of the file within the containing volume"
+    )
+    digest: Digest = pydantic.Field(
+        default_factory=Digest,
+        description="One or more message digests (hashes) of the file data",
+    )
+
+
+class FileStorageURL(DyffSchemaBaseModel):
+    file: File
+    signedURL: StorageSignedURL
+
+
+class Volume(DyffSchemaBaseModel):
+    files: list[File] = pydantic.Field(description="A list of files in the volume")
+
+
 class ModelSourceKinds(str, enum.Enum):
     GitLFS = "GitLFS"
     HuggingFaceHub = "HuggingFaceHub"
@@ -1129,10 +1165,6 @@ class ModelSource(DyffSchemaBaseModel):
     )
 
 
-class ContainerImage(DyffSchemaBaseModel):
-    pass
-
-
 class AcceleratorGPU(DyffSchemaBaseModel):
     hardwareTypes: list[str] = pydantic.Field(
         min_length=1,
@@ -1154,14 +1186,23 @@ class Accelerator(DyffSchemaBaseModel):
     )
 
 
+# FIXME: Remove .storage because it's replaced by ScratchVolume, and rename to
+# ResourceRequirements for consistency with k8s.
 class ModelResources(DyffSchemaBaseModel):
-    storage: Quantity = pydantic.Field(
-        description="Amount of storage required for packaged model, in k8s Quantity notation",
+    storage: Optional[Quantity] = pydantic.Field(
+        default=None,
+        deprecated=True,
+        description="Storage required for packaged model, in k8s Quantity notation",
     )
 
     memory: Optional[Quantity] = pydantic.Field(
         default=None,
-        description="Amount of memory required to run the model on CPU, in k8s Quantity notation",
+        description="Memory, in k8s Quantity notation",
+    )
+
+    cpu: Optional[Quantity] = pydantic.Field(
+        default=None,
+        description="Number of CPUs, in k8s Quantity notation",
     )
 
 
@@ -1175,6 +1216,7 @@ class ModelStorageMedium(str, enum.Enum):
 class ModelArtifactKind(str, enum.Enum):
     HuggingFaceCache = "HuggingFaceCache"
     Mock = "Mock"
+    Volume = "Volume"
 
 
 class ModelArtifactHuggingFaceCache(DyffSchemaBaseModel):
@@ -1194,6 +1236,9 @@ class ModelArtifact(DyffSchemaBaseModel):
     huggingFaceCache: Optional[ModelArtifactHuggingFaceCache] = pydantic.Field(
         default=None, description="Model stored in a HuggingFace cache"
     )
+    volume: Optional[Volume] = pydantic.Field(
+        default=None, description="Model stored as a generic volume"
+    )
 
 
 class ModelStorage(DyffSchemaBaseModel):
@@ -1245,6 +1290,7 @@ class InferenceServiceBuilder(DyffSchemaBaseModel):
 
 class InferenceServiceRunnerKind(str, Enum):
     BENTOML_SERVICE_OPENLLM = "bentoml_service_openllm"
+    CONTAINER = "container"
     HUGGINGFACE = "huggingface"
     MOCK = "mock"
     STANDALONE = "standalone"
@@ -1306,29 +1352,116 @@ class ContainerImageSource(DyffSchemaBaseModel):
         return v
 
 
-class InferenceServiceRunner(DyffSchemaBaseModel):
-    kind: InferenceServiceRunnerKind
+class EnvVar(DyffSchemaBaseModel):
+    name: str
+    value: str
+
+
+# TODO: ModelResources is deprecated
+ResourceRequirements = ModelResources
+
+
+class VolumeMountKind(str, enum.Enum):
+    data = "Data"
+    scratch = "Scratch"
+
+
+class VolumeMountData(DyffSchemaBaseModel):
+    source: EntityIdentifier = pydantic.Field(
+        description="The Dyff platform resource to mount."
+        " Must be a volume-like entity such as a Model or Dataset."
+    )
+
 
+class VolumeMountScratch(DyffSchemaBaseModel):
+    capacity: Quantity = pydantic.Field(
+        description="Minimum storage capacity of the Scratch volume."
+    )
+
+
+class VolumeMount(DyffSchemaBaseModel):
+    kind: VolumeMountKind = pydantic.Field(description="The kind of volume mount.")
+    name: str = pydantic.Field(description="A descriptive name for the mount.")
+    mountPath: Path = pydantic.Field(
+        description="Where to mount the volume in the running container."
+        " Must be an absolute path."
+    )
+    data: Optional[VolumeMountData] = pydantic.Field(
+        default=None, description="Configuration for Data volume mounts."
+    )
+    scratch: Optional[VolumeMountScratch] = pydantic.Field(
+        default=None, description="Configuration for Scratch volume mounts."
+    )
+
+
+class Container(DyffSchemaBaseModel):
+    """Configuration of a runnable container backed by either an image hosted in an
+    external registyr, or an image artifact hosted in the Dyff system.
+
+    This is mostly a subset of the Kubernetes Container schema.
+    """
+
+    args: Optional[list[str]] = pydantic.Field(
+        default=None,
+        description="Arguments to the entrypoint."
+        " The container image's CMD is used if this is not provided.",
+    )
+    command: Optional[list[str]] = pydantic.Field(
+        default=None,
+        description="Entrypoint array. Not executed within a shell."
+        " The container image's ENTRYPOINT is used if this is not provided.",
+    )
+    env: Optional[list[EnvVar]] = pydantic.Field(
+        default=None,
+        description="List of environment variables to set in the container.",
+    )
     # TODO: (DYFF-421) Make .image required
     image: Optional[ContainerImageSource] = pydantic.Field(
         default=None,
-        description="The container image that implements the runner. This field is"
-        " optional for schema backwards-compatibility, but creating new"
-        " services with image=None will result in an error.",
+        description="The container image that implements the runner."
+        " Exactly one of .image and .imageRef must be set.",
     )
-
-    args: Optional[list[str]] = pydantic.Field(
-        default=None, description="Command line arguments to forward to the runner"
+    imageRef: Optional[EntityIdentifier] = pydantic.Field(
+        default=None,
+        description="Container image reference. Must be an image artifact"
+        " that has been uploaded to this Dyff instance and is in Ready status."
+        " Exactly one of .image and .imageRef must be set.",
+    )
+    resources: ResourceRequirements = pydantic.Field(
+        description="Compute Resources required by this container."
+    )
+    volumeMounts: Optional[list[VolumeMount]] = pydantic.Field(
+        default=None, description="Volumes to mount into the container's filesystem."
+    )
+    workingDir: Optional[Path] = pydantic.Field(
+        default=None,
+        description="Container's working directory. If not specified,"
+        " the container runtime's default will be used,"
+        " which might be configured in the container image.",
     )
 
+    @pydantic.model_validator(mode="after")
+    def check_image_exactly_one(cls, values):
+        image = values.image is not None
+        imageRef = values.imageRef is not None
+        if sum([image, imageRef]) != 1:
+            raise ValueError("must specify exactly one of {image, imageRef}")
+        return values
+
+
+class InferenceServiceRunner(Container):
+    """Configuration of a managed runner to use to run an inference service.
+
+    Using a managed runner simplifies the configuration of inference services backed by
+    models in common formats, such as HuggingFace models.
+    """
+
+    kind: InferenceServiceRunnerKind
+
     accelerator: Optional[Accelerator] = pydantic.Field(
         default=None, description="Optional accelerator hardware to use, per node."
     )
 
-    resources: ModelResources = pydantic.Field(
-        description="Resource requirements to run the service, per node."
-    )
-
     nodes: int = pydantic.Field(
         default=1,
         ge=1,
@@ -1366,7 +1499,9 @@ class InferenceServiceBase(DyffSchemaBaseModel):
 
     # FIXME: (DYFF-261) .runner should be required
     runner: Optional[InferenceServiceRunner] = pydantic.Field(
-        default=None, description="Configuration of the Runner used to run the service."
+        default=None,
+        description="Configuration of the managed runner used to run the service."
+        " Only one of .container or .runner may be specified.",
     )
 
     interface: InferenceInterface = pydantic.Field(
@@ -2146,8 +2281,8 @@ class Score(ScoreData):
 
 
 # TODO: (schema-v1) Rename this to Artifact
-class OCIArtifact(Documented, DyffEntity):
-    kind: Literal["Artifact"] = Entities.Artifact.value
+class OCIArtifact(DyffEntity):
+    kind: Literal["OCIArtifact"] = Entities.Artifact.value
 
     manifest: oci.ImageManifest = pydantic.Field(
         description="The OCI image manifest of the artifact"
@@ -2519,6 +2654,7 @@ __all__ = [
     "AuditRequirement",
     "Concern",
     "ConcernBase",
+    "Container",
     "ContainerImageSource",
     "DataSchema",
     "Dataset",
@@ -2552,6 +2688,8 @@ __all__ = [
     "FamilyMemberBase",
     "FamilyMemberKind",
     "FamilyMembers",
+    "File",
+    "FileStorageURL",
     "ForeignInferenceService",
     "ForeignMethod",
     "ForeignModel",
@@ -2616,6 +2754,7 @@ __all__ = [
     "Report",
     "ReportBase",
     "Resources",
+    "ResourceRequirements",
     "Revision",
     "RevisionData",
     "RevisionMetadata",
@@ -2634,6 +2773,11 @@ __all__ = [
     "TagNameType",
     "TaskSchema",
     "UseCase",
+    "Volume",
+    "VolumeMount",
+    "VolumeMountData",
+    "VolumeMountKind",
+    "VolumeMountScratch",
     "entity_class",
     "JobStatus",
     "EntityStatus",

{dyff_schema-0.33.0 → dyff_schema-0.34.0/dyff_schema.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dyff-schema
-Version: 0.33.0
+Version: 0.34.0
 Summary: Data models for the Dyff AI auditing platform.
 Author-email: Digital Safety Research Institute <contact@dsri.org>
 License: Apache-2.0

dyff_schema-0.33.0/dyff/schema/_version.py

@@ -1,2 +0,0 @@
-__version__ = version = "0.33.0"
-__version_tuple__ = version_tuple = (0, 33, 0)