damccorm commented on code in PR #35677:
URL: https://github.com/apache/beam/pull/35677#discussion_r2292012846
##########
sdks/python/apache_beam/ml/transforms/embeddings/vertex_ai.py:
##########
@@ -281,3 +294,194 @@ def get_ptransform_for_processing(self, **kwargs) ->
beam.PTransform:
return RunInference(
model_handler=_ImageEmbeddingHandler(self),
inference_args=self.inference_args)
+
+
+@dataclass
+class VertexAIMultiModalInput:
+ image: Optional[Image] = None
+ video: Optional[Video] = None
+ contextual_text: Optional[str] = None
Review Comment:
> So for something like the mimeType being specifiable that would likely be
something composed in the Image object (if you look at get_embeddings() [from
Vertex](https://cloud.google.com/vertex-ai/generative-ai/docs/reference/python/1.71.0/vertexai.vision_models.MultiModalEmbeddingModel#vertexai_vision_models_MultiModalEmbeddingModel_get_embeddings)
those objects are how we construct the request) and not something we'd have to
explicitly support
Is it part of the Image object today? I agree that is one way it could be
supported, its not clear that it would definitely be the way Vertex would
choose to support it though.
I'm also thinking this would play nicely with Chunks for text (basically we
could have wrappers for all 3 object types which seems fairly natural).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
