Raw Resolver

The raw resolver is a pass-through resolver for pre-computed embedding vectors. It accepts a fixed-length array of floats and returns it unchanged after validating the dimensionality. This is useful when embeddings have been generated externally or by a custom pipeline.

How It Works

The input value is expected to be an array of floats with exactly numDims elements. The resolver validates the length and, if it matches, returns the vector unchanged. No transformation is applied.

Settings

Setting	Type	Default	Description
numDims	integer	—	(required) The expected number of dimensions in the vector.

Embedding Size

The number of dimensions equals the configured numDims value.

Example

{
  "name": "custom_embedding",
  "type": "raw",
  "path": "$.embedding",
  "settings": {
    "numDims": 128
  }
}

When training, documents must contain the embedding as an array of floats at the specified path:

{
  "id": "item-1",
  "doc": {
    "embedding": [
      0.12,
      -0.34,
      0.56,
      ...
    ]
  }
}