corvic-engine 0.3.0rc80__cp38-abi3-win_amd64.whl → 0.3.0rc81__cp38-abi3-win_amd64.whl

corvic/model/__init__.py

@@ -65,6 +65,8 @@ from corvic.model._space import (
     TabularSpace,
     UnknownSpace,
     embedding_model_proto_to_name,
+    image_model_can_embed_images,
+    image_model_can_embed_text,
     image_model_proto_to_name,
 )
 
@@ -120,6 +122,8 @@ __all__ = [
     "add_orm_room_mixin_to_session",
     "embedding_model_proto_to_name",
     "feature_type",
+    "image_model_can_embed_images",
+    "image_model_can_embed_text",
     "image_model_proto_to_name",
     "non_empty_timestamp_to_datetime",
     "space_orm_to_proto",

corvic/model/_space.py

@@ -7,7 +7,7 @@ import copy
 import datetime
 import uuid
 from collections.abc import Iterable, Mapping, Sequence
-from typing import Final, Literal, Self, TypeAlias
+from typing import Final, Literal, Self, TypeAlias, cast
 
 import pyarrow as pa
 import sqlalchemy as sa
@@ -50,14 +50,67 @@ name_to_proto_embedding_model = {
 }
 
 
-image_model_proto_to_name: Final[dict[embedding_models_pb2.ImageModel, str]] = {
-    embedding_models_pb2.IMAGE_MODEL_CUSTOM: "random",
-    embedding_models_pb2.IMAGE_MODEL_CLIP: "openai/clip-vit-base-patch32",
-    embedding_models_pb2.IMAGE_MODEL_IDENTITY: "identity",
-    embedding_models_pb2.IMAGE_MODEL_UNSPECIFIED: "",
-}
+def image_model_proto_to_name(image_model: embedding_models_pb2.ImageModel):
+    match image_model:
+        case embedding_models_pb2.IMAGE_MODEL_CUSTOM:
+            return Ok("random")
+        case embedding_models_pb2.IMAGE_MODEL_CLIP:
+            return Ok("openai/clip-vit-base-patch32")
+        case embedding_models_pb2.IMAGE_MODEL_IDENTITY:
+            return Ok("identity")
+        case embedding_models_pb2.IMAGE_MODEL_SIGLIP2:
+            return Ok("google/siglip2-base-patch16-512")
+        case embedding_models_pb2.IMAGE_MODEL_UNSPECIFIED:
+            return Ok("")
+        case _:
+            return NotFoundError("Could not find image model")
+
+
+def image_model_can_embed_text(image_model: embedding_models_pb2.ImageModel):
+    match image_model:
+        case embedding_models_pb2.IMAGE_MODEL_CUSTOM:
+            return Ok(value=True)
+        case embedding_models_pb2.IMAGE_MODEL_CLIP:
+            return Ok(value=True)
+        case embedding_models_pb2.IMAGE_MODEL_IDENTITY:
+            return Ok(value=True)
+        case embedding_models_pb2.IMAGE_MODEL_SIGLIP2:
+            return Ok(value=True)
+        case embedding_models_pb2.IMAGE_MODEL_UNSPECIFIED:
+            return Ok(value=False)
+        case _:
+            return NotFoundError("Could not find image model")
+
+
+def image_model_can_embed_images(image_model: embedding_models_pb2.ImageModel):
+    match image_model:
+        case embedding_models_pb2.IMAGE_MODEL_CUSTOM:
+            return Ok(value=True)
+        case embedding_models_pb2.IMAGE_MODEL_CLIP:
+            return Ok(value=True)
+        case embedding_models_pb2.IMAGE_MODEL_IDENTITY:
+            return Ok(value=True)
+        case embedding_models_pb2.IMAGE_MODEL_SIGLIP2:
+            return Ok(value=True)
+        case embedding_models_pb2.IMAGE_MODEL_UNSPECIFIED:
+            return Ok(value=False)
+        case _:
+            return NotFoundError("Could not find image model")
+
+
+def _image_model_proto_to_name_unsafe(model: embedding_models_pb2.ImageModel):
+    match image_model_proto_to_name(model):
+        case Ok(value):
+            return value
+        case err:
+            raise err
+
+
 name_to_proto_image_model = {
-    name: model for model, name in image_model_proto_to_name.items()
+    _image_model_proto_to_name_unsafe(
+        cast(embedding_models_pb2.ImageModel, model)
+    ): cast(embedding_models_pb2.ImageModel, model)
+    for model in embedding_models_pb2.ImageModel.values()
 }
 
 
@@ -919,7 +972,7 @@ class EmbedImageParameters:
 
     @property
     def model_name(self) -> str:
-        return image_model_proto_to_name[self.proto_self.model_parameters.model]
+        return _image_model_proto_to_name_unsafe(self.proto_self.model_parameters.model)
 
     @property
     def model(self) -> embedding_models_pb2.ImageModel:

corvic/system/_embedder.py

@@ -1,3 +1,4 @@
+import abc
 import asyncio
 import dataclasses
 from collections.abc import Sequence
@@ -11,10 +12,8 @@ from corvic import eorm
 from corvic.result import InternalError, InvalidArgumentError, Ok
 
 if TYPE_CHECKING:
-    from transformers.models.clip import (
-        CLIPModel,
-        CLIPProcessor,
-    )
+    from transformers.models.auto.modeling_auto import AutoModel
+    from transformers.models.auto.processing_auto import AutoProcessor
 
 
 @dataclasses.dataclass
@@ -84,42 +83,16 @@ class ImageEmbedder(Protocol):
 
 
 @dataclasses.dataclass
-class ClipModels:
-    model: "CLIPModel"
-    processor: "CLIPProcessor"
+class LoadedModels:
+    model: "AutoModel"
+    processor: "AutoProcessor"
 
 
-class ClipText(TextEmbedder):
-    """Clip Text embedder.
-
-    CLIP (Contrastive Language-Image Pre-Training) is a neural network trained
-    on a variety of (image, text) pairs. It can be instructed in natural language
-    to predict the most relevant text snippet, given an image, without
-    directly optimizing for the task, similarly to the zero-shot capabilities of
-    GPT-2 and 3. We found CLIP matches the performance of the original ResNet50
-    on ImageNet "zero-shot" without using any of the original 1.28M labeled examples,
-    overcoming several major challenges in computer vision.
-    """
-
-    def _load_models(self):
-        from transformers.models.clip import (
-            CLIPModel,
-            CLIPProcessor,
-        )
+class HFModelText(TextEmbedder):
+    """Generic text/image embedder from hugging face models."""
 
-        model = CLIPModel.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
-            pretrained_model_name_or_path="openai/clip-vit-base-patch32",
-            revision="5812e510083bb2d23fa43778a39ac065d205ed4d",
-        )
-        processor = cast(
-            CLIPProcessor,
-            CLIPProcessor.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
-                pretrained_model_name_or_path="openai/clip-vit-base-patch32",
-                revision="5812e510083bb2d23fa43778a39ac065d205ed4d",
-                use_fast=False,
-            ),
-        )
-        return ClipModels(model=model, processor=processor)
+    @abc.abstractmethod
+    def _load_models(self) -> LoadedModels: ...
 
     def embed(
         self, context: EmbedTextContext
@@ -133,20 +106,20 @@ class ClipText(TextEmbedder):
         models = self._load_models()
         model = models.model
         processor = models.processor
-        model.eval()
+        model.eval()  # type: ignore[reportAttributeAccess]
 
         import torch
 
         with torch.no_grad():
             inputs = cast(
                 dict[str, torch.Tensor],
-                processor(
+                processor(  # type: ignore[reportAttributeAccess]
                     text=context.inputs,
                     return_tensors="pt",
                     padding=True,
                 ),
             )
-            text_features = model.get_text_features(input_ids=inputs["input_ids"])
+            text_features = model.get_text_features(input_ids=inputs["input_ids"])  # type: ignore[reportAttributeAccess]
 
         text_features_numpy = cast(np.ndarray[Any, Any], text_features.numpy())  #  pyright: ignore[reportUnknownMemberType]
 
@@ -170,3 +143,65 @@ class ClipText(TextEmbedder):
         return await asyncio.get_running_loop().run_in_executor(
             worker_threads, self.embed, context
         )
+
+
+class ClipText(HFModelText):
+    """Clip Text embedder.
+
+    CLIP (Contrastive Language-Image Pre-Training) is a neural network trained
+    on a variety of (image, text) pairs. It can be instructed in natural language
+    to predict the most relevant text snippet, given an image, without
+    directly optimizing for the task, similarly to the zero-shot capabilities of
+    GPT-2 and 3. We found CLIP matches the performance of the original ResNet50
+    on ImageNet "zero-shot" without using any of the original 1.28M labeled examples,
+    overcoming several major challenges in computer vision.
+    """
+
+    def _load_models(self):
+        from transformers.models.clip import (
+            CLIPModel,
+            CLIPProcessor,
+        )
+
+        model = cast(
+            AutoModel,
+            CLIPModel.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
+                pretrained_model_name_or_path="openai/clip-vit-base-patch32",
+                revision="5812e510083bb2d23fa43778a39ac065d205ed4d",
+            ),
+        )
+        processor = cast(
+            AutoProcessor,
+            CLIPProcessor.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
+                pretrained_model_name_or_path="openai/clip-vit-base-patch32",
+                revision="5812e510083bb2d23fa43778a39ac065d205ed4d",
+                use_fast=False,
+            ),
+        )
+        return LoadedModels(model=model, processor=processor)
+
+
+class SigLIP2Text(HFModelText):
+    """SigLIP2 text/image embedder."""
+
+    def _load_models(self):
+        from transformers.models.auto.modeling_auto import AutoModel
+        from transformers.models.auto.processing_auto import AutoProcessor
+
+        model = cast(
+            AutoModel,
+            AutoModel.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
+                pretrained_model_name_or_path="google/siglip2-base-patch16-512",
+                revision="a89f5c5093f902bf39d3cd4d81d2c09867f0724b",
+                device_map="auto",
+            ),
+        )
+        processor = cast(
+            AutoProcessor,
+            AutoProcessor.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
+                pretrained_model_name_or_path="google/siglip2-base-patch16-512",
+                revision="a89f5c5093f902bf39d3cd4d81d2c09867f0724b",
+                use_fast=True,
+            ),
+        )
+        return LoadedModels(model=model, processor=processor)

corvic/system/_image_embedder.py

@@ -1,3 +1,4 @@
+import abc
 import asyncio
 import dataclasses
 from concurrent.futures import ThreadPoolExecutor
@@ -16,10 +17,8 @@ from corvic.system._embedder import (
 
 if TYPE_CHECKING:
     from PIL import Image
-    from transformers.models.clip import (
-        CLIPModel,
-        CLIPProcessor,
-    )
+    from transformers.models.auto.modeling_auto import AutoModel
+    from transformers.models.auto.processing_auto import AutoProcessor
 
 
 class RandomImageEmbedder(ImageEmbedder):
@@ -75,42 +74,16 @@ def image_from_bytes(
 
 
 @dataclasses.dataclass
-class ClipModels:
-    model: "CLIPModel"
-    processor: "CLIPProcessor"
+class LoadedModels:
+    model: "AutoModel"
+    processor: "AutoProcessor"
 
 
-class Clip(ImageEmbedder):
-    """Clip image embedder.
-
-    CLIP (Contrastive Language-Image Pre-Training) is a neural network trained
-    on a variety of (image, text) pairs. It can be instructed in natural language
-    to predict the most relevant text snippet, given an image, without
-    directly optimizing for the task, similarly to the zero-shot capabilities of
-    GPT-2 and 3. We found CLIP matches the performance of the original ResNet50
-    on ImageNet “zero-shot” without using any of the original 1.28M labeled examples,
-    overcoming several major challenges in computer vision.
-    """
+class HFModelImageEmbedder(ImageEmbedder):
+    """Generic image embedder from hugging face models."""
 
-    def _load_models(self):
-        from transformers.models.clip import (
-            CLIPModel,
-            CLIPProcessor,
-        )
-
-        model = CLIPModel.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
-            pretrained_model_name_or_path="openai/clip-vit-base-patch32",
-            revision="5812e510083bb2d23fa43778a39ac065d205ed4d",
-        )
-        processor = cast(
-            CLIPProcessor,
-            CLIPProcessor.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
-                pretrained_model_name_or_path="openai/clip-vit-base-patch32",
-                revision="5812e510083bb2d23fa43778a39ac065d205ed4d",
-                use_fast=False,
-            ),
-        )
-        return ClipModels(model=model, processor=processor)
+    @abc.abstractmethod
+    def _load_models(self) -> LoadedModels: ...
 
     def embed(
         self, context: EmbedImageContext
@@ -144,16 +117,16 @@ class Clip(ImageEmbedder):
         models = self._load_models()
         model = models.model
         processor = models.processor
-        model.eval()
+        model.eval()  # type: ignore[reportAttributeAccess]
 
         import torch
 
         with torch.no_grad():
             inputs = cast(
                 dict[str, torch.FloatTensor],
-                processor(images=images, return_tensors="pt"),
+                processor(images=images, return_tensors="pt"),  # type: ignore[reportAttributeAccess]
             )
-            image_features = model.get_image_features(
+            image_features = model.get_image_features(  # type: ignore[reportAttributeAccess]
                 pixel_values=inputs["pixel_values"]
             )
 
@@ -180,19 +153,51 @@ class Clip(ImageEmbedder):
         )
 
 
-class SigLIP2(Clip):
-    """SigLIP2 image embedder."""
+class Clip(HFModelImageEmbedder):
+    """Clip image embedder.
 
-    def _load_models(self):
-        from transformers.models.auto.modeling_auto import AutoModel
-        from transformers.models.auto.processing_auto import AutoProcessor
+    CLIP (Contrastive Language-Image Pre-Training) is a neural network trained
+    on a variety of (image, text) pairs. It can be instructed in natural language
+    to predict the most relevant text snippet, given an image, without
+    directly optimizing for the task, similarly to the zero-shot capabilities of
+    GPT-2 and 3. We found CLIP matches the performance of the original ResNet50
+    on ImageNet “zero-shot” without using any of the original 1.28M labeled examples,
+    overcoming several major challenges in computer vision.
+    """
+
+    def _load_models(self) -> LoadedModels:
         from transformers.models.clip import (
             CLIPModel,
             CLIPProcessor,
         )
 
         model = cast(
-            CLIPModel,
+            AutoModel,
+            CLIPModel.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
+                pretrained_model_name_or_path="openai/clip-vit-base-patch32",
+                revision="5812e510083bb2d23fa43778a39ac065d205ed4d",
+            ),
+        )
+        processor = cast(
+            AutoProcessor,
+            CLIPProcessor.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
+                pretrained_model_name_or_path="openai/clip-vit-base-patch32",
+                revision="5812e510083bb2d23fa43778a39ac065d205ed4d",
+                use_fast=False,
+            ),
+        )
+        return LoadedModels(model=model, processor=processor)
+
+
+class SigLIP2(HFModelImageEmbedder):
+    """SigLIP2 image embedder."""
+
+    def _load_models(self):
+        from transformers.models.auto.modeling_auto import AutoModel
+        from transformers.models.auto.processing_auto import AutoProcessor
+
+        model = cast(
+            AutoModel,
             AutoModel.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
                 pretrained_model_name_or_path="google/siglip2-base-patch16-512",
                 revision="a89f5c5093f902bf39d3cd4d81d2c09867f0724b",
@@ -200,13 +205,14 @@ class SigLIP2(Clip):
             ),
         )
         processor = cast(
-            CLIPProcessor,
+            AutoProcessor,
             AutoProcessor.from_pretrained(  # pyright: ignore[reportUnknownMemberType]
                 pretrained_model_name_or_path="google/siglip2-base-patch16-512",
                 revision="a89f5c5093f902bf39d3cd4d81d2c09867f0724b",
+                use_fast=True,
             ),
         )
-        return ClipModels(model=model, processor=processor)
+        return LoadedModels(model=model, processor=processor)
 
 
 class CombinedImageEmbedder(ImageEmbedder):

{corvic_engine-0.3.0rc80.dist-info → corvic_engine-0.3.0rc81.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: corvic-engine
-Version: 0.3.0rc80
+Version: 0.3.0rc81
 Classifier: Environment :: Console
 Classifier: License :: Other/Proprietary License
 Classifier: Programming Language :: Python :: Implementation :: CPython

{corvic_engine-0.3.0rc80.dist-info → corvic_engine-0.3.0rc81.dist-info}/RECORD

@@ -11,11 +11,11 @@ corvic/embedding_metric/__init__.py,sha256=8a-QKSQNbiksueHk5LdkugjZr6wasP4ff8A-J
 corvic/embedding_metric/embeddings.py,sha256=XCiMzoGdRSmCOJnBDnxm3xlU0L_vrXwUxEjwdMv1FMI,14036
 corvic/embedding_metric/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 corvic/engine/__init__.py,sha256=XL4Vg7rNcBi29ccVelpeFizR9oJtGYXDn84W9zok9d4,975
-corvic/engine/_native.pyd,sha256=V8crGQ6LIdFBAZ_TZYIZbv_ULgLHuF8Kll7xPX7ELGE,438272
+corvic/engine/_native.pyd,sha256=qr5bmOjrt1JfUr_JrvlhPY8Cz3GcqJ_REeAchdUnKm4,438272
 corvic/engine/_native.pyi,sha256=KYMPtvXqHZ-jMgZohLf4se3rr-rBpCihmjANcr6s8ag,1390
 corvic/engine/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 corvic/eorm/__init__.py,sha256=b4dFnu4fW7wj3Y0SMNVXOp8KoKOp_HAL4GyDN-S8fOY,13704
-corvic/model/__init__.py,sha256=ZtMq8LTJE013v-OoVq0muo0aqomNGrjmXwKfS8wd4CU,3075
+corvic/model/__init__.py,sha256=umu7rhilpzsKgeGHfNnq1bFf3vRk5gECSOSNkRVfdfc,3215
 corvic/model/_base_model.py,sha256=BBcpX792AC8zb1_Jq_aFQ7KwB3H5Mn4z1NE9sAoReqA,10328
 corvic/model/_completion_model.py,sha256=e6vGrON3NTc9vXweunJ4hp9byMvMd-igMzaul-genK4,7643
 corvic/model/_defaults.py,sha256=OnROutSYhCuNuIvJbWdp4NZyamYtWRjwc4BE2ZqNAm8,1529
@@ -27,7 +27,7 @@ corvic/model/_proto_orm_convert.py,sha256=czEE6qYjg4s73IMSA_mGYAyh9s2FKc2ofjE6YV
 corvic/model/_resource.py,sha256=3eeYJf8M8063Go0QVOM03QqRAshyChInqqdpSxRx8mQ,8519
 corvic/model/_room.py,sha256=OqceFrRZVt4OcHHbhAgCj0s77ErmIcaXRsf8Lv0nfrM,2868
 corvic/model/_source.py,sha256=f62oLSv3qdOmfVhEjqL5ie55YiDdGDq0IhSAfFvrkBw,9803
-corvic/model/_space.py,sha256=ffEIPWXTtuAp9kBA7J45h95OMwe0vFUlb-Hbot40AC0,36497
+corvic/model/_space.py,sha256=uBBi6H5VIg1z-r4DXOPqYb4DUznGkv2a8_YGr6wx3zo,38505
 corvic/model/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 corvic/op_graph/__init__.py,sha256=1DMrQfuuS3FkLa9DXYDjSDLurdxxpG5H1jB2ctaa9xo,1444
 corvic/op_graph/_schema.py,sha256=rDZt6xAFpvhBZlp54uvXlqTqOxO8uFtTtukBGSUylqA,5688
@@ -71,8 +71,8 @@ corvic/sql/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 corvic/system/__init__.py,sha256=Qddpk3SoHgoiBJyr9w2bPkp477NZv5MDR8701A17nCY,2697
 corvic/system/_column_encoding.py,sha256=feSWIv4vKstVq-aavWPk53YucUiq7rZvuyofqTicXBE,7574
 corvic/system/_dimension_reduction.py,sha256=2tg5SIHY4P480DJQj6PSjW1VgAJCAVJAH8D3BY-ZYXA,2964
-corvic/system/_embedder.py,sha256=Jr-f4rwNdFRZuMaW-prPCDtjNkkNZVn7zCGMNi_hEYw,5424
-corvic/system/_image_embedder.py,sha256=jTuSjxwh8n1peSidKCQFV_LLJDH4XelX9NV4uvGBBtw,11829
+corvic/system/_embedder.py,sha256=WJC4RtPLoPDDIgquKhpmPmYi2bAoR5_oVkWyBAepJTE,6864
+corvic/system/_image_embedder.py,sha256=Af3MI3VkiZQxCQjgDwaCNNpgEsnAiqfa4bdET63bvss,12218
 corvic/system/_planner.py,sha256=ecL-HW8PVz5eWJ1Ktf-RAD2IdZkHu3GuBtXdqElo4ts,8210
 corvic/system/_text_embedder.py,sha256=NDi--3_tzwIWVImjhFWmp8dHmydGGXNu6GYH8qODsIc,4000
 corvic/system/client.py,sha256=JcA-fPraqDkl9f8BiClS0qeGY6wzKcEDPymutWrJo54,812
@@ -94,9 +94,9 @@ corvic/version/__init__.py,sha256=JlkRLvKXsu3zIxhdynO_0Ub5NfQOvGjfwCRkNnaOu9U,11
 corvic/version/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 corvic/well_known_types/__init__.py,sha256=Btbeqieik2AcmijeOXeqBptzueBpgNitvH9J5VNm12w,1289
 corvic/well_known_types/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-corvic_engine-0.3.0rc80.dist-info/METADATA,sha256=h5FHy0y1OwCZu1dkPk1gY1Fb60l5z6wMe5DV9qR3aH8,1814
-corvic_engine-0.3.0rc80.dist-info/WHEEL,sha256=qo08K5WTt1v9liGoFGXfI182ciKs5521XAErJtzFynQ,94
-corvic_engine-0.3.0rc80.dist-info/licenses/LICENSE,sha256=DSS1OD0oIgssKOmAzkMRBv5jvvVuZQbrIv8lpl9DXY8,1035
+corvic_engine-0.3.0rc81.dist-info/METADATA,sha256=xgNR-_243a1DK7ug0D9mgjiwstj5ht4OBfduUojkwMI,1814
+corvic_engine-0.3.0rc81.dist-info/WHEEL,sha256=qo08K5WTt1v9liGoFGXfI182ciKs5521XAErJtzFynQ,94
+corvic_engine-0.3.0rc81.dist-info/licenses/LICENSE,sha256=DSS1OD0oIgssKOmAzkMRBv5jvvVuZQbrIv8lpl9DXY8,1035
 corvic_generated/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 corvic_generated/algorithm/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 corvic_generated/algorithm/graph/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -107,8 +107,8 @@ corvic_generated/algorithm/graph/v1/graph_pb2_grpc.py,sha256=_bXoS025FcWrXR1E_3M
 corvic_generated/algorithm/graph/v1/graph_pb2_grpc.pyi,sha256=H9-ADaiKR9iyVZvmnXutZqWwRRCDxjUIktkfJrJFIHg,417
 corvic_generated/embedding/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 corvic_generated/embedding/v1/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-corvic_generated/embedding/v1/models_pb2.py,sha256=uQDda8455N2roSi7gkFED5BqPzjw2RNDgtuSBDhJmac,4433
-corvic_generated/embedding/v1/models_pb2.pyi,sha256=yM8zBLxeARMolH1lRJLO4hNKWKAHF0xsygAKmMufBj0,4270
+corvic_generated/embedding/v1/models_pb2.py,sha256=i0gLpcb4LLBE11C4MIMFgWaM0fdwq9rFnYFVOmuWHiY,4481
+corvic_generated/embedding/v1/models_pb2.pyi,sha256=NzQ1TuhJ_IKp2xYjdf1qxu9VgwyMza7WPcKMneI-Ycg,4351
 corvic_generated/embedding/v1/models_pb2_grpc.py,sha256=_bXoS025FcWrXR1E_3Mh4GHB1RMvgz8lIpit-Awnf-s,163
 corvic_generated/embedding/v1/models_pb2_grpc.pyi,sha256=H9-ADaiKR9iyVZvmnXutZqWwRRCDxjUIktkfJrJFIHg,417
 corvic_generated/feature/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -207,4 +207,4 @@ corvic_generated/status/v1/service_pb2.py,sha256=CKXPX2ahq8O4cFhPpt6wo6l--6VZcgj
 corvic_generated/status/v1/service_pb2.pyi,sha256=iXLR2FOKQJpBgvBzpD2kVwcYOCksP2aRwK4JYaI9CBw,558
 corvic_generated/status/v1/service_pb2_grpc.py,sha256=y-a5ldrphWlNJW-yKswyjNmXokK4-5bbEEfczjagJHo,2736
 corvic_generated/status/v1/service_pb2_grpc.pyi,sha256=OoAnaZ64FD0UTzPoRhYvQU8ecoilhHj3ySjSfHbVDaU,1501
-corvic_engine-0.3.0rc80.dist-info/RECORD,,
+corvic_engine-0.3.0rc81.dist-info/RECORD,,

corvic_generated/embedding/v1/models_pb2.py

@@ -14,7 +14,7 @@ _sym_db = _symbol_database.Default()
 
 
 
-DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n corvic/embedding/v1/models.proto\x12\x13\x63orvic.embedding.v1\"R\n\nParameters\x12\x30\n\x05model\x18\x01 \x01(\x0e\x32\x1a.corvic.embedding.v1.ModelR\x05model\x12\x12\n\x04ndim\x18\x02 \x01(\x05R\x04ndim\"\xf4\x01\n\x19\x43olumnEmbeddingParameters\x12q\n\x11\x63olumn_parameters\x18\x01 \x03(\x0b\x32\x44.corvic.embedding.v1.ColumnEmbeddingParameters.ColumnParametersEntryR\x10\x63olumnParameters\x1a\x64\n\x15\x43olumnParametersEntry\x12\x10\n\x03key\x18\x01 \x01(\tR\x03key\x12\x35\n\x05value\x18\x02 \x01(\x0b\x32\x1f.corvic.embedding.v1.ParametersR\x05value:\x02\x38\x01\"\x8f\x01\n\x1e\x43oncatStringAndEmbedParameters\x12!\n\x0c\x63olumn_names\x18\x01 \x03(\tR\x0b\x63olumnNames\x12J\n\x10model_parameters\x18\x02 \x01(\x0b\x32\x1f.corvic.embedding.v1.ParametersR\x0fmodelParameters\"\x89\x01\n\x18\x43oncatAndEmbedParameters\x12!\n\x0c\x63olumn_names\x18\x01 \x03(\tR\x0b\x63olumnNames\x12J\n\x10model_parameters\x18\x02 \x01(\x0b\x32\x1f.corvic.embedding.v1.ParametersR\x0fmodelParameters\".\n\x18\x45mbedAndConcatParameters\x12\x12\n\x04ndim\x18\x01 \x01(\x05R\x04ndim\"a\n\x14ImageModelParameters\x12\x35\n\x05model\x18\x01 \x01(\x0e\x32\x1f.corvic.embedding.v1.ImageModelR\x05model\x12\x12\n\x04ndim\x18\x02 \x01(\x05R\x04ndim\"\x8d\x01\n\x14\x45mbedImageParameters\x12\x1f\n\x0b\x63olumn_name\x18\x01 \x01(\tR\ncolumnName\x12T\n\x10model_parameters\x18\x02 \x01(\x0b\x32).corvic.embedding.v1.ImageModelParametersR\x0fmodelParameters*\xdc\x01\n\x05Model\x12\x15\n\x11MODEL_UNSPECIFIED\x10\x00\x12\"\n\x1aMODEL_SENTENCE_TRANSFORMER\x10\x01\x1a\x02\x08\x01\x12\'\n#MODEL_OPENAI_TEXT_EMBEDDING_3_SMALL\x10\x03\x12\'\n#MODEL_OPENAI_TEXT_EMBEDDING_3_LARGE\x10\x04\x12 \n\x1cMODEL_GCP_TEXT_EMBEDDING_004\x10\x05\x12\x10\n\x0cMODEL_CUSTOM\x10\x02\x12\x12\n\x0eMODEL_IDENTITY\x10\x06*q\n\nImageModel\x12\x1b\n\x17IMAGE_MODEL_UNSPECIFIED\x10\x00\x12\x14\n\x10IMAGE_MODEL_CLIP\x10\x01\x12\x16\n\x12IMAGE_MODEL_CUSTOM\x10\x02\x12\x18\n\x14IMAGE_MODEL_IDENTITY\x10\x03\x62\x06proto3')
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n corvic/embedding/v1/models.proto\x12\x13\x63orvic.embedding.v1\"R\n\nParameters\x12\x30\n\x05model\x18\x01 \x01(\x0e\x32\x1a.corvic.embedding.v1.ModelR\x05model\x12\x12\n\x04ndim\x18\x02 \x01(\x05R\x04ndim\"\xf4\x01\n\x19\x43olumnEmbeddingParameters\x12q\n\x11\x63olumn_parameters\x18\x01 \x03(\x0b\x32\x44.corvic.embedding.v1.ColumnEmbeddingParameters.ColumnParametersEntryR\x10\x63olumnParameters\x1a\x64\n\x15\x43olumnParametersEntry\x12\x10\n\x03key\x18\x01 \x01(\tR\x03key\x12\x35\n\x05value\x18\x02 \x01(\x0b\x32\x1f.corvic.embedding.v1.ParametersR\x05value:\x02\x38\x01\"\x8f\x01\n\x1e\x43oncatStringAndEmbedParameters\x12!\n\x0c\x63olumn_names\x18\x01 \x03(\tR\x0b\x63olumnNames\x12J\n\x10model_parameters\x18\x02 \x01(\x0b\x32\x1f.corvic.embedding.v1.ParametersR\x0fmodelParameters\"\x89\x01\n\x18\x43oncatAndEmbedParameters\x12!\n\x0c\x63olumn_names\x18\x01 \x03(\tR\x0b\x63olumnNames\x12J\n\x10model_parameters\x18\x02 \x01(\x0b\x32\x1f.corvic.embedding.v1.ParametersR\x0fmodelParameters\".\n\x18\x45mbedAndConcatParameters\x12\x12\n\x04ndim\x18\x01 \x01(\x05R\x04ndim\"a\n\x14ImageModelParameters\x12\x35\n\x05model\x18\x01 \x01(\x0e\x32\x1f.corvic.embedding.v1.ImageModelR\x05model\x12\x12\n\x04ndim\x18\x02 \x01(\x05R\x04ndim\"\x8d\x01\n\x14\x45mbedImageParameters\x12\x1f\n\x0b\x63olumn_name\x18\x01 \x01(\tR\ncolumnName\x12T\n\x10model_parameters\x18\x02 \x01(\x0b\x32).corvic.embedding.v1.ImageModelParametersR\x0fmodelParameters*\xdc\x01\n\x05Model\x12\x15\n\x11MODEL_UNSPECIFIED\x10\x00\x12\"\n\x1aMODEL_SENTENCE_TRANSFORMER\x10\x01\x1a\x02\x08\x01\x12\'\n#MODEL_OPENAI_TEXT_EMBEDDING_3_SMALL\x10\x03\x12\'\n#MODEL_OPENAI_TEXT_EMBEDDING_3_LARGE\x10\x04\x12 \n\x1cMODEL_GCP_TEXT_EMBEDDING_004\x10\x05\x12\x10\n\x0cMODEL_CUSTOM\x10\x02\x12\x12\n\x0eMODEL_IDENTITY\x10\x06*\x8a\x01\n\nImageModel\x12\x1b\n\x17IMAGE_MODEL_UNSPECIFIED\x10\x00\x12\x14\n\x10IMAGE_MODEL_CLIP\x10\x01\x12\x16\n\x12IMAGE_MODEL_CUSTOM\x10\x02\x12\x18\n\x14IMAGE_MODEL_IDENTITY\x10\x03\x12\x17\n\x13IMAGE_MODEL_SIGLIP2\x10\x04\x62\x06proto3')
 
 _globals = globals()
 _builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
@@ -27,8 +27,8 @@ if _descriptor._USE_C_DESCRIPTORS == False:
   _globals['_COLUMNEMBEDDINGPARAMETERS_COLUMNPARAMETERSENTRY']._serialized_options = b'8\001'
   _globals['_MODEL']._serialized_start=966
   _globals['_MODEL']._serialized_end=1186
-  _globals['_IMAGEMODEL']._serialized_start=1188
-  _globals['_IMAGEMODEL']._serialized_end=1301
+  _globals['_IMAGEMODEL']._serialized_start=1189
+  _globals['_IMAGEMODEL']._serialized_end=1327
   _globals['_PARAMETERS']._serialized_start=57
   _globals['_PARAMETERS']._serialized_end=139
   _globals['_COLUMNEMBEDDINGPARAMETERS']._serialized_start=142

corvic_generated/embedding/v1/models_pb2.pyi

@@ -22,6 +22,7 @@ class ImageModel(int, metaclass=_enum_type_wrapper.EnumTypeWrapper):
     IMAGE_MODEL_CLIP: _ClassVar[ImageModel]
     IMAGE_MODEL_CUSTOM: _ClassVar[ImageModel]
     IMAGE_MODEL_IDENTITY: _ClassVar[ImageModel]
+    IMAGE_MODEL_SIGLIP2: _ClassVar[ImageModel]
 MODEL_UNSPECIFIED: Model
 MODEL_SENTENCE_TRANSFORMER: Model
 MODEL_OPENAI_TEXT_EMBEDDING_3_SMALL: Model
@@ -33,6 +34,7 @@ IMAGE_MODEL_UNSPECIFIED: ImageModel
 IMAGE_MODEL_CLIP: ImageModel
 IMAGE_MODEL_CUSTOM: ImageModel
 IMAGE_MODEL_IDENTITY: ImageModel
+IMAGE_MODEL_SIGLIP2: ImageModel
 
 class Parameters(_message.Message):
     __slots__ = ("model", "ndim")