Merge pull request #726 from bioimage-io/dev

add weights.onnx.external_data

Author: FynnBe

changelog.md

@@ -10,6 +10,10 @@ In this file we log both:
 
 This changelog includes implementation details and my reference the [changes to the Resource Description Format](#changes-to-the-resource-description-format), e.g. in entry [bioimageio.spec 0.5.2](#bioimageiospec-052).
 
+#### bioimageio.spec 0.5.6.0
+
+- [model format 0.5.6](#model-056)
+
 #### bioimageio.spec 0.5.5.7
 
 - Fix `ValidationSummary.add_detail()` (respect 'valid-format' status)
@@ -42,7 +46,7 @@ This changelog includes implementation details and my reference the [changes to
 
 #### bioimageio.spec 0.5.5.0
 
-- added programmatic `uplaod` (alpha stage)
+- added programmatic `upload` (alpha stage)
 - [generic format 0.3.2 and model format 0.5.5](#generic-032-and-model-055)
 
 #### bioimageio.spec 0.5.4.4
@@ -330,6 +334,11 @@ Here is a list of changes for each `type` and `format_version`.
 Note that 'generic' changes apply to `type` application, dataset and notebook of the same `format_versions`.
 If the changes also apply to `type` model, the coresponding model format version is noted, e.g. [generic 0.3.1 and model 0.5.3](#generic-031-and-model-053).
 
+#### model 0.5.6
+
+- Non-breaking changes:
+  - New optional `weights.onnx.external_data` field to allow for (large) ONNX models split into an architecture and an external data (weights) file.
+
 #### generic 0.3.2 and model 0.5.5
 
 - Non-breaking changes:

src/bioimageio/spec/_version.py

@@ -1 +1 @@
-VERSION = "0.5.5.7"
+VERSION = "0.5.6.0"

src/bioimageio/spec/model/v0_5.py

@@ -2359,12 +2359,23 @@ class KerasHdf5WeightsDescr(WeightsEntryDescrBase):
     """TensorFlow version used to create these weights."""
 
 
+FileDescr_external_data = Annotated[
+    FileDescr_,
+    WithSuffix(".data", case_sensitive=True),
+    Field(examples=[dict(source="weights.onnx.data")]),
+]
+
+
 class OnnxWeightsDescr(WeightsEntryDescrBase):
     type = "onnx"
     weights_format_name: ClassVar[str] = "ONNX"
     opset_version: Annotated[int, Ge(7)]
     """ONNX opset version"""
 
+    external_data: Optional[FileDescr_external_data] = None
+    """Source of the external ONNX data file holding the weights.
+    (If present **source** holds the ONNX architecture without weights)."""
+
 
 class PytorchStateDictWeightsDescr(WeightsEntryDescrBase):
     type = "pytorch_state_dict"
@@ -2611,11 +2622,11 @@ class ModelDescr(GenericModelDescrBase):
     These fields are typically stored in a YAML file which we call a model resource description file (model RDF).
     """
 
-    implemented_format_version: ClassVar[Literal["0.5.5"]] = "0.5.5"
+    implemented_format_version: ClassVar[Literal["0.5.6"]] = "0.5.6"
     if TYPE_CHECKING:
-        format_version: Literal["0.5.5"] = "0.5.5"
+        format_version: Literal["0.5.6"] = "0.5.6"
     else:
-        format_version: Literal["0.5.5"]
+        format_version: Literal["0.5.6"]
         """Version of the bioimage.io model description specification used.
         When creating a new model always use the latest micro/patch version described here.
         The `format_version` is important for any consumer software to understand how to parse the fields.
@@ -3357,7 +3368,7 @@ def conv_authors(auths: Optional[Sequence[_Author_v0_4]]):
             covers=src.covers,
             description=src.description,
             documentation=src.documentation,
-            format_version="0.5.5",
+            format_version="0.5.6",
             git_repo=src.git_repo,  # pyright: ignore[reportArgumentType]
             icon=src.icon,
             id=None if src.id is None else ModelId(src.id),

src/bioimageio/spec/static/bioimageio_schema.json

@@ -10528,7 +10528,7 @@
           "title": "Version Comment"
         },
         "format_version": {
-          "const": "0.5.5",
+          "const": "0.5.6",
           "description": "Version of the bioimage.io model description specification used.\nWhen creating a new model always use the latest micro/patch version described here.\nThe `format_version` is important for any consumer software to understand how to parse the fields.",
           "title": "Format Version",
           "type": "string"
@@ -10677,7 +10677,7 @@
         "outputs",
         "weights"
       ],
-      "title": "model 0.5.5",
+      "title": "model 0.5.6",
       "type": "object"
     },
     "bioimageio__spec__model__v0_5__OnnxWeightsDescr": {
@@ -10773,6 +10773,23 @@
           "minimum": 7,
           "title": "Opset Version",
           "type": "integer"
+        },
+        "external_data": {
+          "anyOf": [
+            {
+              "$ref": "#/$defs/FileDescr",
+              "examples": [
+                {
+                  "source": "weights.onnx.data"
+                }
+              ]
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "default": null,
+          "description": "Source of the external ONNX data file holding the weights.\n(If present **source** holds the ONNX architecture without weights)."
         }
       },
       "required": [
@@ -13892,7 +13909,7 @@
         "discriminator": {
           "mapping": {
             "0.4.10": "#/$defs/bioimageio__spec__model__v0_4__ModelDescr",
-            "0.5.5": "#/$defs/bioimageio__spec__model__v0_5__ModelDescr"
+            "0.5.6": "#/$defs/bioimageio__spec__model__v0_5__ModelDescr"
           },
           "propertyName": "format_version"
         },
@@ -13976,7 +13993,7 @@
       "discriminator": {
         "mapping": {
           "0.4.10": "#/$defs/bioimageio__spec__model__v0_4__ModelDescr",
-          "0.5.5": "#/$defs/bioimageio__spec__model__v0_5__ModelDescr"
+          "0.5.6": "#/$defs/bioimageio__spec__model__v0_5__ModelDescr"
         },
         "propertyName": "format_version"
       },