Merge branch 'main' into suffic

llama_stack/providers/utils/inference/litellm_openai_mixin.py

@@ -4,8 +4,6 @@
 # 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 AsyncGenerator, AsyncIterator
 from typing import Any
 
@@ -37,7 +35,6 @@ from llama_stack.apis.inference.inference import (
     OpenAIChatCompletion,
     OpenAIChatCompletionChunk,
     OpenAICompletion,
-    OpenAIEmbeddingData,
     OpenAIEmbeddingsResponse,
     OpenAIEmbeddingUsage,
     OpenAIMessageParam,
@@ -48,6 +45,7 @@ from llama_stack.distribution.request_headers import NeedsRequestProviderData
 from llama_stack.log import get_logger
 from llama_stack.providers.utils.inference.model_registry import ModelRegistryHelper
 from llama_stack.providers.utils.inference.openai_compat import (
+    b64_encode_openai_embeddings_response,
     convert_message_to_openai_dict_new,
     convert_openai_chat_completion_choice,
     convert_openai_chat_completion_stream,
@@ -293,16 +291,7 @@ class LiteLLMOpenAIMixin(
         )
 
         # Convert response to OpenAI format
-        data = []
-        for i, embedding_data in enumerate(response["data"]):
-            # we encode to base64 if the encoding format is base64 in the request
-            if encoding_format == "base64":
-                byte_data = b"".join(struct.pack("f", f) for f in embedding_data["embedding"])
-                embedding = base64.b64encode(byte_data).decode("utf-8")
-            else:
-                embedding = embedding_data["embedding"]
-
-            data.append(OpenAIEmbeddingData(embedding=embedding, index=i))
+        data = b64_encode_openai_embeddings_response(response.data, encoding_format)
 
         usage = OpenAIEmbeddingUsage(
             prompt_tokens=response["usage"]["prompt_tokens"],

llama_stack/providers/utils/inference/openai_compat.py

@@ -3,8 +3,10 @@
 #
 # 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 logging
+import struct
 import time
 import uuid
 import warnings
@@ -108,6 +110,7 @@ from llama_stack.apis.inference.inference import (
     OpenAIChatCompletion,
     OpenAICompletion,
     OpenAICompletionChoice,
+    OpenAIEmbeddingData,
     OpenAIMessageParam,
     OpenAIResponseFormatParam,
     ToolConfig,
@@ -1484,3 +1487,55 @@ class OpenAIChatCompletionToLlamaStackMixin:
             model=model,
             object="chat.completion",
         )
+
+
+def prepare_openai_embeddings_params(
+    model: str,
+    input: str | list[str],
+    encoding_format: str | None = "float",
+    dimensions: int | None = None,
+    user: str | None = None,
+):
+    if model is None:
+        raise ValueError("Model must be provided for embeddings")
+
+    input_list = [input] if isinstance(input, str) else input
+
+    params: dict[str, Any] = {
+        "model": model,
+        "input": input_list,
+    }
+
+    if encoding_format is not None:
+        params["encoding_format"] = encoding_format
+    if dimensions is not None:
+        params["dimensions"] = dimensions
+    if user is not None:
+        params["user"] = user
+
+    return params
+
+
+def b64_encode_openai_embeddings_response(
+    response_data: 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