Hello,
Happy to see the new image embedding using Siglip2 available in Pipeline Builder.
I could not find any documentation (too early?) on it. Is there any plan to support this function natively in Code Repository as well ? Maybe with transforms.mediasets?
Note: It would be good if Palantir could be more transparent on the upcoming functionalities like these to avoid working on logics that would be available few weeks later natively !
Cheers,
Hi!
Happy to see organic usage of the new Image to Embeddings board already! You are right to point out no public documentation yet. We are planning to ship similar functionality on the querying side of the ontology and then announce an image semantic search package.
For the code repositories part - yes, it should be available as a generateEmbedding transform already!
Let us know if you have any questions.
Sorry, would you mind sharing where this function is located ?
I have looked into palantir_models.models and palantir_models.transforms but I could only find what I believe is the text embedding one (“GenericEmbeddingModel”) which are relying on language_model_service_api.
Here is the new public documentation for the method in Code Repositories: https://www.palantir.com/docs/foundry/transforms-python/media-set-transforms-api#generate_image_embeddings