[Feature]: About non-symmetric embedding support

[ZaynJarvis]

Problem Statement

we are seeing many PR supporting non-symmetrics embedding
#608 #624 #635 #636 etc.

Proposed Solution

My suggestion for now is

1. don't give config options
2. carefully check embedding context is right (query_embedder used in query scenario only, not miss used)
3. set context in embedder or embedding request contexual_embedder(context) -> XXXEmbedder(context)
4. for Embedder that supports non-symmetric, map context to static value of [type, input_type, task etc.]. if changing it is required, use envvar, don't bloat config file with provider specific settings.

Alternatives Considered

No response

Feature Area

Model Integration

Use Case

to make long term (task_type [query / document]) support for embedding models: Jina, OpenAI, Minimax, Gemini etc.

Example API (Optional)

> set in request level, is embedder is used in both query & document context
            return OpenAIDenseEmbedder(
                model_name=self.dense.model,
                api_key=self.dense.api_key,
                api_base=self.dense.api_base,
                dimension=self.dense.dimension,
                context=context,
            )
            ...
            return JinaDenseEmbedder(
                model_name=self.dense.model,
                api_key=self.dense.api_key,
                api_base=self.dense.api_base,
                dimension=self.dense.dimension,
                context=context,
            )
---

in XXXEmbedder impl:

  class OpenAIDenseEmbedder:
    ...
    input_type = "x" if context == "query" else "y"


  class JinaDenseEmbedder:
    ...
    task  = "a" if context == "query" else "b"


would this be better?

Additional Context

No response

Contribution

• I am willing to contribute to implementing this feature