Address review comments

Signed-off-by: Bill Murdock <bmurdock@redhat.com>

llama_stack/providers/registry/inference.py

@@ -277,7 +277,7 @@ Available Models:
             api=Api.inference,
             adapter_type="watsonx",
             provider_type="remote::watsonx",
-            pip_packages=["ibm_watsonx_ai"],
+            pip_packages=["litellm"],
             module="llama_stack.providers.remote.inference.watsonx",
             config_class="llama_stack.providers.remote.inference.watsonx.WatsonXConfig",
             provider_data_validator="llama_stack.providers.remote.inference.watsonx.WatsonXProviderDataValidator",

llama_stack/providers/remote/inference/watsonx/config.py

@@ -25,12 +25,16 @@ class WatsonXConfig(RemoteInferenceProviderConfig):
         default_factory=lambda: os.getenv("WATSONX_BASE_URL", "https://us-south.ml.cloud.ibm.com"),
         description="A base url for accessing the watsonx.ai",
     )
+    # This seems like it should be required, but none of the other remote inference
+    # providers require it, so this is optional here too for consistency.
+    # The OpenAIConfig uses default=None instead, so this is following that precedent.
     api_key: SecretStr | None = Field(
-        default_factory=lambda: os.getenv("WATSONX_API_KEY"),
+        default=None,
         description="The watsonx.ai API key",
     )
+    # As above, this is optional here too for consistency.
     project_id: str | None = Field(
-        default_factory=lambda: os.getenv("WATSONX_PROJECT_ID"),
+        default=None,
         description="The watsonx.ai project ID",
     )
     timeout: int = Field(

llama_stack/providers/remote/inference/watsonx/watsonx.py

@@ -16,9 +16,6 @@ from llama_stack.providers.utils.inference.litellm_openai_mixin import LiteLLMOp
 
 
 class WatsonXInferenceAdapter(LiteLLMOpenAIMixin):
-    _config: WatsonXConfig
-    __provider_id__: str = "watsonx"
-
     def __init__(self, config: WatsonXConfig):
         LiteLLMOpenAIMixin.__init__(
             self,
@@ -29,17 +26,9 @@ class WatsonXInferenceAdapter(LiteLLMOpenAIMixin):
         self.available_models = None
         self.config = config
 
-    # get_api_key = LiteLLMOpenAIMixin.get_api_key
-
     def get_base_url(self) -> str:
         return self.config.url
 
-    async def initialize(self):
-        await super().initialize()
-
-    async def shutdown(self):
-        await super().shutdown()
-
     async def _get_params(self, request: ChatCompletionRequest) -> dict[str, Any]:
         # Get base parameters from parent
         params = await super()._get_params(request)

llama_stack/providers/utils/inference/litellm_openai_mixin.py

@@ -4,6 +4,8 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import base64
+import struct
 from collections.abc import AsyncIterator
 from typing import Any
 
@@ -16,6 +18,7 @@ from llama_stack.apis.inference import (
     OpenAIChatCompletion,
     OpenAIChatCompletionChunk,
     OpenAICompletion,
+    OpenAIEmbeddingData,
     OpenAIEmbeddingsResponse,
     OpenAIEmbeddingUsage,
     OpenAIMessageParam,
@@ -26,7 +29,6 @@ from llama_stack.core.request_headers import NeedsRequestProviderData
 from llama_stack.log import get_logger
 from llama_stack.providers.utils.inference.model_registry import ModelRegistryHelper, ProviderModelEntry
 from llama_stack.providers.utils.inference.openai_compat import (
-    b64_encode_openai_embeddings_response,
     convert_message_to_openai_dict_new,
     convert_tooldef_to_openai_tool,
     get_sampling_options,
@@ -334,6 +336,7 @@ class LiteLLMOpenAIMixin(
             api_key=self.get_api_key(),
             api_base=self.api_base,
         )
+        logger.info(f"params to litellm (openai compat): {params}")
         return await litellm.acompletion(**params)
 
     async def check_model_availability(self, model: str) -> bool:
@@ -349,3 +352,28 @@ class LiteLLMOpenAIMixin(
             return False
 
         return model in litellm.models_by_provider[self.litellm_provider_name]
+
+
+def b64_encode_openai_embeddings_response(
+    response_data: list[dict], encoding_format: str | None = "float"
+) -> list[OpenAIEmbeddingData]:
+    """
+    Process the OpenAI embeddings response to encode the embeddings in base64 format if specified.
+    """
+    data = []
+    for i, embedding_data in enumerate(response_data):
+        if encoding_format == "base64":
+            byte_array = bytearray()
+            for embedding_value in embedding_data["embedding"]:
+                byte_array.extend(struct.pack("f", float(embedding_value)))
+
+            response_embedding = base64.b64encode(byte_array).decode("utf-8")
+        else:
+            response_embedding = embedding_data["embedding"]
+        data.append(
+            OpenAIEmbeddingData(
+                embedding=response_embedding,
+                index=i,
+            )
+        )
+    return data

llama_stack/providers/utils/inference/openai_compat.py

@@ -3,9 +3,7 @@
 #
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
-import base64
 import json
-import struct
 import time
 import uuid
 import warnings
@@ -103,7 +101,6 @@ from llama_stack.apis.inference import (
     JsonSchemaResponseFormat,
     Message,
     OpenAIChatCompletion,
-    OpenAIEmbeddingData,
     OpenAIMessageParam,
     OpenAIResponseFormatParam,
     SamplingParams,
@@ -1402,28 +1399,3 @@ def prepare_openai_embeddings_params(
         params["user"] = user
 
     return params
-
-
-def b64_encode_openai_embeddings_response(
-    response_data: list[dict], encoding_format: str | None = "float"
-) -> list[OpenAIEmbeddingData]:
-    """
-    Process the OpenAI embeddings response to encode the embeddings in base64 format if specified.
-    """
-    data = []
-    for i, embedding_data in enumerate(response_data):
-        if encoding_format == "base64":
-            byte_array = bytearray()
-            for embedding_value in embedding_data["embedding"]:
-                byte_array.extend(struct.pack("f", float(embedding_value)))
-
-            response_embedding = base64.b64encode(byte_array).decode("utf-8")
-        else:
-            response_embedding = embedding_data["embedding"]
-        data.append(
-            OpenAIEmbeddingData(
-                embedding=response_embedding,
-                index=i,
-            )
-        )
-    return data