feat(api) expose resolved model package metadata in HTTP inference responses

inference/core/entities/responses/action_recognition.py

@@ -1,13 +1,14 @@
 from typing import List
 
-from pydantic import BaseModel, Field
+from pydantic import Field
 
+from inference.core.entities.responses.inference import InferenceResponse
 from inference.core.workflows.core_steps.models.roboflow.action_recognition.entities import (  # noqa: F401
     ActionRecognitionPrediction,
 )
 
 
-class ActionRecognitionInferenceResponse(BaseModel):
+class ActionRecognitionInferenceResponse(InferenceResponse):
     """Classified ranges covering one clip.
 
     Frame indices count from the first frame of the submitted clip, so a

inference/core/entities/responses/inference.py

@@ -10,6 +10,7 @@
     InferenceResponse,
     InferenceResponseImage,
     InstanceSegmentationInferenceResponse,
+    ResolvedModel,
     WithVisualizationResponse,
 )
 from inference.core.workflows.core_steps.common.segmentation_entities import (  # noqa: F401
@@ -232,6 +233,7 @@ class InstanceSegmentationInferenceResponseDC:
     frame_id: object = None
     time: object = None
     visualization: object = None
+    resolved_model: Optional[ResolvedModel] = None
     # Internal stream-pipeline fast path: lets workflow execution carry a
     # response future through Model.infer_from_request without blocking the
     # inference thread. `_is_response_dc_to_dict` intentionally ignores it.
@@ -283,6 +285,8 @@ def _is_response_dc_to_dict(r: InstanceSegmentationInferenceResponseDC) -> dict:
         d["time"] = r.time
     if r.visualization is not None:
         d["visualization"] = r.visualization
+    if r.resolved_model is not None:
+        d["resolved_model"] = r.resolved_model.model_dump()
     return d
 
 
@@ -362,7 +366,7 @@ class FaceDetectionPrediction(ObjectDetectionPrediction):
     landmarks: Union[List[Point], List[Point3D]]
 
 
-class DepthEstimationResponse(BaseModel):
+class DepthEstimationResponse(InferenceResponse):
     """Response for depth estimation inference.
 
     Attributes:

inference/core/entities/responses/ocr.py

@@ -1,14 +1,15 @@
 from typing import List, Optional
 
-from pydantic import BaseModel, Field
+from pydantic import Field
 
 from inference.core.entities.responses.inference import (
+    InferenceResponse,
     InferenceResponseImage,
     ObjectDetectionPrediction,
 )
 
 
-class OCRInferenceResponse(BaseModel):
+class OCRInferenceResponse(InferenceResponse):
     """
     OCR Inference response.
 

inference/core/entities/responses/sam.py

@@ -1,9 +1,11 @@
 from typing import Any, List, Union
 
-from pydantic import BaseModel, Field
+from pydantic import Field
 
+from inference.core.entities.responses.inference import InferenceResponse
 
-class SamEmbeddingResponse(BaseModel):
+
+class SamEmbeddingResponse(InferenceResponse):
     """SAM embedding response.
 
     Attributes:
@@ -20,7 +22,7 @@ class SamEmbeddingResponse(BaseModel):
     )
 
 
-class SamSegmentationResponse(BaseModel):
+class SamSegmentationResponse(InferenceResponse):
     """SAM segmentation response.
 
     Attributes:

inference/core/entities/responses/sam2.py

@@ -1,13 +1,14 @@
 from typing import List
 
-from pydantic import BaseModel, Field
+from pydantic import Field
 
+from inference.core.entities.responses.inference import InferenceResponse
 from inference.core.workflows.core_steps.common.segmentation_entities import (  # noqa: F401
     Sam2SegmentationPrediction,
 )
 
 
-class Sam2EmbeddingResponse(BaseModel):
+class Sam2EmbeddingResponse(InferenceResponse):
     """SAM embedding response.
 
     Attributes:
@@ -21,7 +22,7 @@ class Sam2EmbeddingResponse(BaseModel):
     )
 
 
-class Sam2SegmentationResponse(BaseModel):
+class Sam2SegmentationResponse(InferenceResponse):
     predictions: List[Sam2SegmentationPrediction] = Field()
     time: float = Field(
         description="The time in seconds it took to produce the segmentation including preprocessing"

inference/core/entities/responses/sam3.py

@@ -2,8 +2,10 @@
 
 from pydantic import BaseModel, Field
 
+from inference.core.entities.responses.inference import InferenceResponse
 
-class Sam3EmbeddingResponse(BaseModel):
+
+class Sam3EmbeddingResponse(InferenceResponse):
     image_id: str = Field(description="Image id embeddings are cached to")
     time: float = Field(
         description="The time in seconds it took to produce the embeddings including preprocessing"
@@ -33,7 +35,7 @@ class Sam3PromptResult(BaseModel):
     predictions: List[Sam3SegmentationPrediction] = Field()
 
 
-class Sam3SegmentationResponse(BaseModel):
+class Sam3SegmentationResponse(InferenceResponse):
     prompt_results: List[Sam3PromptResult] = Field()
     time: float = Field(
         description="The time in seconds it took to produce the segmentation including preprocessing"

inference/core/interfaces/http/http_api.py

@@ -4256,6 +4256,7 @@ def _infer_depth_estimation(
                             depth_data["normalized_depth"]
                         )
                     return DepthEstimationResponse(
+                        resolved_model=getattr(response, "resolved_model", None),
                         normalized_depth=serialized_depth,
                         depth_map_format=inference_request.depth_map_format,
                         image=depth_data["image"].base64_image,

inference/core/models/base.py

@@ -5,7 +5,11 @@
 
 from inference.core import logger
 from inference.core.entities.requests.inference import InferenceRequest
-from inference.core.entities.responses.inference import InferenceResponse
+from inference.core.entities.responses.inference import (
+    InferenceResponse,
+    InstanceSegmentationInferenceResponseDC,
+    ResolvedModel,
+)
 from inference.core.env import USE_INFERENCE_MODELS
 from inference.core.models.types import PreprocessReturnMetadata
 from inference.core.telemetry import set_span_attribute, start_span
@@ -106,6 +110,31 @@ class Model(BaseInference):
         clear_cache(): Clears any cache if necessary.
     """
 
+    @property
+    def resolved_model(self) -> Any:
+        model = getattr(self, "_model", None)
+        if USE_INFERENCE_MODELS and model is not None:
+            return getattr(model, "resolved_model", None)
+        model_id = getattr(self, "endpoint", None) or getattr(self, "model_id", None)
+        return ResolvedModel(model_id=model_id) if model_id is not None else None
+
+    def _attach_resolved_model_metadata(self, responses: Any) -> None:
+        metadata = self.resolved_model
+        if metadata is None:
+            return
+        resolved_model = ResolvedModel(
+            model_id=metadata.model_id,
+            model_package_id=metadata.model_package_id,
+            backend=metadata.backend,
+            quantization=metadata.quantization,
+        )
+        responses = responses if isinstance(responses, list) else [responses]
+        for response in responses:
+            if isinstance(
+                response, (InferenceResponse, InstanceSegmentationInferenceResponseDC)
+            ):
+                response.resolved_model = resolved_model
+
     def log(self, m):
         """Prints the given message.
 
@@ -181,6 +210,7 @@ def infer_from_request(
         if not isinstance(request.image, list) and len(responses) > 0:
             responses = responses[0]
 
+        self._attach_resolved_model_metadata(responses)
         return responses
 
     def make_response(

inference/core/models/classification_base.py

@@ -297,6 +297,7 @@ def infer_from_request(
         if not isinstance(request.image, list):
             responses = responses[0]
 
+        self._attach_resolved_model_metadata(responses)
         return responses
 
     def make_response(

inference/core/models/inference_models_adapters.py

@@ -1508,6 +1508,7 @@ def infer_from_request(
         if not isinstance(request.image, list):
             responses = responses[0]
 
+        self._attach_resolved_model_metadata(responses)
         return responses
 
 
@@ -2040,12 +2041,14 @@ def infer_from_request(
                     stride=max(1.0, source_fps / window.sample_fps),
                 )
         timeline.sort(key=lambda entry: (entry.start_frame_idx, entry.class_id))
-        return ActionRecognitionInferenceResponse(
+        response = ActionRecognitionInferenceResponse(
             timeline=timeline,
             source_fps=source_fps,
             frame_count=frame_count,
             windows_classified=windows_classified,
         )
+        self._attach_resolved_model_metadata(response)
+        return response
 
     def preprocess(self, *args, **kwargs):
         raise NotImplementedError(

inference/core/workflows/__init__.py

@@ -4,6 +4,7 @@
 installs the compat finder and rebinds this module in ``sys.modules`` to the
 canonical package so the two names share one object.
 """
+
 from inference._workflows_compat import alias_legacy_root as _alias_legacy_root
 
 _alias_legacy_root(__name__)

inference/enterprise/workflows/__init__.py

@@ -6,6 +6,7 @@
 but it installs the compat finder so child imports of ``enterprise_blocks``
 resolve to canonical modules.
 """
+
 from inference._workflows_compat import install as _install
 
 _install()

inference/models/clip/clip_inference_models.py

@@ -343,6 +343,7 @@ def infer_from_request(
         data = infer_func(**request.dict())
         response = make_response_func(data)
         response.time = perf_counter() - t1
+        self._attach_resolved_model_metadata(response)
         return response
 
     def make_response(self, embeddings, *args, **kwargs) -> InferenceResponse:

inference/models/clip/clip_model.py

@@ -335,6 +335,7 @@ def infer_from_request(
         data = infer_func(**request.dict())
         response = make_response_func(data)
         response.time = perf_counter() - t1
+        self._attach_resolved_model_metadata(response)
         return response
 
     def make_response(self, embeddings, *args, **kwargs) -> InferenceResponse:

inference/models/doctr/doctr_model.py

@@ -104,8 +104,11 @@ def infer_from_request(
             for image in request.image:
                 request_copy.image = image
                 response.append(self.single_request(request=request_copy))
+            self._attach_resolved_model_metadata(response)
             return response
-        return self.single_request(request)
+        response = self.single_request(request)
+        self._attach_resolved_model_metadata(response)
+        return response
 
     def single_request(self, request: DoctrOCRInferenceRequest) -> OCRInferenceResponse:
         t1 = perf_counter()

inference/models/doctr/doctr_model_inference_models.py

@@ -82,12 +82,14 @@ def infer_from_request(
     ) -> Union[OCRInferenceResponse, List]:
         if type(request.image) is list:
             response = []
-            request_copy = copy.copy(request)
+            request_copy = copy(request)
             for image in request.image:
                 request_copy.image = image
                 response.append(self.single_request(request=request_copy))
-            return response
-        return self.single_request(request)
+        else:
+            response = self.single_request(request)
+        self._attach_resolved_model_metadata(response)
+        return response
 
     def single_request(self, request: DoctrOCRInferenceRequest) -> OCRInferenceResponse:
         t1 = perf_counter()

inference/models/easy_ocr/easy_ocr.py

@@ -127,8 +127,11 @@ def infer_from_request(
             for image in request.image:
                 request_copy.image = image
                 response.append(self.single_request(request=request_copy))
+            self._attach_resolved_model_metadata(response)
             return response
-        return self.single_request(request)
+        response = self.single_request(request)
+        self._attach_resolved_model_metadata(response)
+        return response
 
     def single_request(self, request: EasyOCRInferenceRequest) -> OCRInferenceResponse:
         t1 = perf_counter()

inference/models/easy_ocr/easy_ocr_inference_models.py

@@ -102,8 +102,10 @@ def infer_from_request(
             for image in request.image:
                 request_copy.image = image
                 response.append(self.single_request(request=request_copy))
-            return response
-        return self.single_request(request)
+        else:
+            response = self.single_request(request)
+        self._attach_resolved_model_metadata(response)
+        return response
 
     def single_request(self, request: EasyOCRInferenceRequest) -> OCRInferenceResponse:
         t1 = perf_counter()

inference/models/grounding_dino/grounding_dino.py

@@ -151,6 +151,7 @@ def infer_from_request(
         Perform inference based on the details provided in the request, and return the associated responses.
         """
         result = self.infer(**request.dict())
+        self._attach_resolved_model_metadata(result)
         return result
 
     @usage_collector("model")

inference/models/grounding_dino/grounding_dino_inference_models.py

@@ -95,6 +95,7 @@ def infer_from_request(
         Perform inference based on the details provided in the request, and return the associated responses.
         """
         result = self.infer(**request.dict())
+        self._attach_resolved_model_metadata(result)
         return result
 
     @usage_collector("model")

inference/models/perception_encoder/perception_encoder.py

@@ -301,6 +301,7 @@ def infer_from_request(
         data = infer_func(**request.dict())
         response = make_response_func(data)
         response.time = perf_counter() - t1
+        self._attach_resolved_model_metadata(response)
         return response
 
     def make_response(self, embeddings, *args, **kwargs) -> InferenceResponse:

inference/models/perception_encoder/perception_encoder_inference_models.py

@@ -287,6 +287,7 @@ def infer_from_request(
         data = infer_func(**request.dict())
         response = make_response_func(data)
         response.time = perf_counter() - t1
+        self._attach_resolved_model_metadata(response)
         return response
 
     def make_response(self, embeddings, *args, **kwargs) -> InferenceResponse:

inference/models/sam/segment_anything.py

@@ -141,16 +141,20 @@ def infer_from_request(self, request: SamInferenceRequest):
                 embedding, _ = self.embed_image(**request.dict())
                 inference_time = perf_counter() - t1
                 if request.format == "json":
-                    return SamEmbeddingResponse(
+                    response = SamEmbeddingResponse(
                         embeddings=embedding.tolist(), time=inference_time
                     )
+                    self._attach_resolved_model_metadata(response)
+                    return response
                 elif request.format == "binary":
                     binary_vector = BytesIO()
                     np.save(binary_vector, embedding)
                     binary_vector.seek(0)
-                    return SamEmbeddingResponse(
+                    response = SamEmbeddingResponse(
                         embeddings=binary_vector.getvalue(), time=inference_time
                     )
+                    self._attach_resolved_model_metadata(response)
+                    return response
             elif isinstance(request, SamSegmentationRequest):
                 masks, low_res_masks = self.segment_image(**request.dict())
                 if request.format == "json":
@@ -174,6 +178,7 @@ def infer_from_request(self, request: SamInferenceRequest):
                     low_res_masks=[m.tolist() for m in low_res_masks],
                     time=perf_counter() - t1,
                 )
+                self._attach_resolved_model_metadata(response)
                 return response
 
     def preproc_image(self, image: InferenceRequestImage):