feat: Adding OCI Embeddings (#4300)

# What does this PR do?
Enabling usage of OCI embedding models.

## Test Plan
Testing embedding model:
`OCI_COMPARTMENT_OCID="" OCI_REGION="us-chicago-1"
OCI_AUTH_TYPE=config_file pytest -sv
tests/integration/inference/test_openai_embeddings.py --stack-config oci
--embedding-model oci/openai.text-embedding-3-small --inference-mode
live`

Testing chat model:
`OCI_COMPARTMENT_OCID="" OCI_REGION="us-chicago-1"
OCI_AUTH_TYPE=config_file pytest -sv tests/integration/inference/
--stack-config oci --text-model oci/openai.gpt-4.1-nano-2025-04-14
--inference-mode live`

Testing curl for embeddings:
`curl -X POST http://localhost:8321/v1/embeddings -H "Content-Type:
application/json" -d '{
  "model": "oci/openai.text-embedding-3-small",
  "input": ["First text", "Second text"],
  "encoding_format": "float"
}'`

`{"object":"list","data":[{"object":"embedding","embedding":[-0.017190756...0.025272394],"index":1}],"model":"oci/openai.text-embedding-3-small","usage":{"prompt_tokens":4,"total_tokens":4}}`

---------

Co-authored-by: Omar Abdelwahab <omaryashraf10@gmail.com>

src/llama_stack/providers/remote/inference/oci/oci.py

@@ -18,11 +18,7 @@ from llama_stack.log import get_logger
 from llama_stack.providers.remote.inference.oci.auth import OciInstancePrincipalAuth, OciUserPrincipalAuth
 from llama_stack.providers.remote.inference.oci.config import OCIConfig
 from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
-from llama_stack_api import (
-    ModelType,
-    OpenAIEmbeddingsRequestWithExtraBody,
-    OpenAIEmbeddingsResponse,
-)
+from llama_stack_api import Model, ModelType
 
 logger = get_logger(name=__name__, category="inference::oci")
 
@@ -37,6 +33,8 @@ MODEL_CAPABILITIES = ["TEXT_GENERATION", "TEXT_SUMMARIZATION", "TEXT_EMBEDDINGS"
 class OCIInferenceAdapter(OpenAIMixin):
     config: OCIConfig
 
+    embedding_models: list[str] = []
+
     async def initialize(self) -> None:
         """Initialize and validate OCI configuration."""
         if self.config.oci_auth_type not in VALID_OCI_AUTH_TYPES:
@@ -113,7 +111,9 @@ class OCIInferenceAdapter(OpenAIMixin):
             client = GenerativeAiClient(config=oci_config, signer=oci_signer)
 
         models: ModelCollection = client.list_models(
-            compartment_id=compartment_id, capability=MODEL_CAPABILITIES, lifecycle_state="ACTIVE"
+            compartment_id=compartment_id,
+            # capability=MODEL_CAPABILITIES,
+            lifecycle_state="ACTIVE",
         ).data
 
         seen_models = set()
@@ -122,7 +122,7 @@ class OCIInferenceAdapter(OpenAIMixin):
             if model.time_deprecated or model.time_on_demand_retired:
                 continue
 
-            if "CHAT" not in model.capabilities or "FINE_TUNE" in model.capabilities:
+            if "UNKNOWN_ENUM_VALUE" in model.capabilities or "FINE_TUNE" in model.capabilities:
                 continue
 
             # Use display_name + model_type as the key to avoid conflicts
@@ -133,8 +133,30 @@ class OCIInferenceAdapter(OpenAIMixin):
             seen_models.add(model_key)
             model_ids.append(model.display_name)
 
+            if "TEXT_EMBEDDINGS" in model.capabilities:
+                self.embedding_models.append(model.display_name)
+
         return model_ids
 
-    async def openai_embeddings(self, params: OpenAIEmbeddingsRequestWithExtraBody) -> OpenAIEmbeddingsResponse:
-        # The constructed url is a mask that hits OCI's "chat" action, which is not supported for embeddings.
-        raise NotImplementedError("OCI Provider does not (currently) support embeddings")
+    def construct_model_from_identifier(self, identifier: str) -> Model:
+        """
+        Construct a Model instance corresponding to the given identifier
+
+        Child classes can override this to customize model typing/metadata.
+
+        :param identifier: The provider's model identifier
+        :return: A Model instance
+        """
+        if identifier in self.embedding_models:
+            return Model(
+                provider_id=self.__provider_id__,  # type: ignore[attr-defined]
+                provider_resource_id=identifier,
+                identifier=identifier,
+                model_type=ModelType.embedding,
+            )
+        return Model(
+            provider_id=self.__provider_id__,  # type: ignore[attr-defined]
+            provider_resource_id=identifier,
+            identifier=identifier,
+            model_type=ModelType.llm,
+        )

tests/integration/inference/test_openai_embeddings.py

@@ -138,7 +138,6 @@ def skip_if_model_doesnt_support_openai_embeddings(client, model_id):
         "remote::runpod",
         "remote::sambanova",
         "remote::tgi",
-        "remote::oci",
     ):
         pytest.skip(f"Model {model_id} hosted by {provider.provider_type} doesn't support OpenAI embeddings.")