I want to explore using special markup tags, like , in text inputs to embedding models in order to encourage the model to perform intermediate reasoning before producing embeddings.
Specifically:
How can I structure input text so that content inside is interpreted as “reasoning steps” rather than direct content?
Is there a way to make embedding models leverage this internal reasoning to produce richer or more context-aware embeddings?
Are there existing practices or techniques for “pre-reasoning” with embedding models using inline markers or tags?
I want to explore using special markup tags, like , in text inputs to embedding models in order to encourage the model to perform intermediate reasoning before producing embeddings.
Specifically:
How can I structure input text so that content inside is interpreted as “reasoning steps” rather than direct content?
Is there a way to make embedding models leverage this internal reasoning to produce richer or more context-aware embeddings?
Are there existing practices or techniques for “pre-reasoning” with embedding models using inline markers or tags?