Home Schema Example JSON Schema

Embedding Model Settings

Type: object

Defines settings for the embedding model during data and query flow.

root default_prompt_name

Prompt Name

Type: enum (of string) Default: "query"

Prompt for the embedding.

Must be one of:

  • "query"
  • "text"

root dimension

Dimension

Type: integer Default: -1

Dimension of the embedding

root emb_model_source

Model Source

Type: enum (of string) Default: "huggingface"

Where is the model from?

Must be one of:

  • "huggingface"
  • "bedrock"

root hnsw_kwargs

HNSW settings

Default: null

HNSW settings

root hnsw_kwargs anyOf HNSW Graph Settings

HNSW Graph Settings

Type: object

Defines settings for HNSW graph. See
https://github.com/run-llama/llamaindex/blob/977d60a058c691957dae3eb3c66c1894faea24ac/llama-index-integrations/vectorstores/llama-index-vector-stores-postgres/llamaindex/vectorstores/postgres/base.py#L570

root hnsw_kwargs anyOf HNSW Graph Settings hnsw_dist_method

dist_method

Type: enum (of string) Default: "vector_cosine_ops"

Distance metric to use. Note that by default PGVectorStore.buildquery calls cosine_distance

Must be one of:

  • "vector_l2_ops"
  • "vector_ip_ops"
  • "vector_cosine_ops"
  • "vector_l1_ops"
  • "bit_hamming_ops"
  • "bit_jaccard_ops"

root hnsw_kwargs anyOf HNSW Graph Settings hnsw_ef_construction

ef_construction

Type: integer Default: 64

Size of the dynamic candidate list for constructing the graph. Higher value provides better recall at the cost of speed

root hnsw_kwargs anyOf HNSW Graph Settings hnsw_m

m

Type: integer Default: 16

Max number of connections per layer.
root hnsw_kwargs anyOf item 1
Type: null

root model

Model

Type: string

Name of embedding model to use.

Must be at least 1 characters long

root query_instruction

Query Instruction

Type: string Default: "Represent this sentence for searching relevant passages:"

Instruction for the query.

root context_size

Context Size

Type: integer Default: 512

Context size of the embedding model