Error adding skillset to Azure Search Indexer

Question

I am trying to add a custom skillset to my indexer. I am however facing some issues which I don't know how to solve. My skills:

# https://learn.microsoft.com/en-us/python/api/azure-search-documents/azure.search.documents.indexes.models.splitskill?view=azure-python
split_skill = SplitSkill(
    name="split",
    description=None,
    context="/document/reviews_text",
    inputs=[{"name": "text", "source": "/document/content"}],
    outputs=[{"name": "textItems", "targetName": "pages"}],
    text_split_mode="pages",
    default_language_code="en",
    maximum_page_length=1000
)

# https://learn.microsoft.com/en-us/python/api/azure-search-documents/azure.search.documents.indexes.models.azureopenaiembeddingskill?view=azure-python-preview
text_embedding_skill = AzureOpenAIEmbeddingSkill(
    name="embedding",
    description=None,
    inputs=[{"name": "text", "source": "/pages"}],
    outputs=[{"name": "embeddings", "targetName": "embeddings"}],
    resource_uri="xxx",
    deployment_id="yyy",
    api_key="zzz",
)

How they are added to an index

# Define the skillset with the text embedding skill
skillset = SearchIndexerSkillset(
    name="my-text-embedding-skillset",
    skills=[split_skill, text_embedding_skill],
    description="A skillset for creating text embeddings"
)

The error I get is:

azure.core.exceptions.HttpResponseError: () One or more skills are invalid. Details: Error in skill 'embedding': Outputs are not supported by skill: embeddings
Code: 
Message: One or more skills are invalid. Details: Error in skill 'embedding': Outputs are not supported by skill: embeddings

Should I create the index first and match the output name to the one in the index? I would be grateful if anybody could point to a good tutorial on this :)

Answer 1

The name in the outputs for the AzureOpenAIEmbeddingSkill needs to be the singular embedding, as documented here. You can make the targetName whatever you want (to use in later skills/mapping to the index), but the name needs to be what is documented.