def get_embedding()

in experiments/legacy/backend/embeddings.py [0:0]

• 36 lines of code
• 12 McCabe index (conditional complexity)

  def get_embedding(self, text: Optional[str] = None, 
                    image: Optional[str] = None, base64: bool = False):
    """Invoke Vertex multimodal embedding API.
    
    You can pass text and/or image. If neither is passed will raise exception

    Args:
      text: text to embed
      image: can be local file path, GCS URI or base64 encoded image
      base64: True indicates image is base64. False (default) will be 
        interpreted as image path (either local or GCS)
    Returns:
    named tuple with the following attributes:
      text_embedding: 1408 dimension vector of type Sequence[float]
      image_embedding: 1408 dimension vector of type Sequence[float] OR None if
        no image provide
    """
    if not text and not image:
      raise ValueError('At least one of text or image_bytes must be specified.')

    instance = struct_pb2.Struct()
    if text:
      if len(text) >= 1024:
        logging.warning('Text must be less than 1024 characters. Truncating text.')
        text = text[:1023]
      instance.fields['text'].string_value = text

    if image:
      image_struct = instance.fields['image'].struct_value
      if base64:
        image_struct.fields['bytesBase64Encoded'].string_value = image
      elif image.lower().startswith('gs://'):
        image_struct.fields['gcsUri'].string_value = image
      else:
        with open(image, "rb") as f:
          image_bytes = f.read()
        encoded_content = base64.b64encode(image_bytes).decode("utf-8")
        image_struct.fields['bytesBase64Encoded'].string_value = encoded_content

    instances = [instance]
    endpoint = (f"projects/{self.project}/locations/{self.location}"
      "/publishers/google/models/multimodalembedding@001")
    response = self.client.predict(endpoint=endpoint, instances=instances)

    text_embedding = None
    if text:
      text_emb_value = response.predictions[0]['textEmbedding']
      text_embedding = [v for v in text_emb_value]

    image_embedding = None
    if image:
      image_emb_value = response.predictions[0]['imageEmbedding']
      image_embedding = [v for v in image_emb_value]

    return EmbeddingResponse(
      text_embedding=text_embedding,
      image_embedding=image_embedding)