158 lines
4.6 KiB
Python
158 lines
4.6 KiB
Python
import math
|
|
from typing import Any, List
|
|
|
|
from llama_index.schema import BaseNode, MetadataMode, TextNode
|
|
from llama_index.vector_stores.types import (
|
|
MetadataFilters,
|
|
VectorStore,
|
|
VectorStoreQuery,
|
|
VectorStoreQueryResult,
|
|
)
|
|
from llama_index.vector_stores.utils import (
|
|
legacy_metadata_dict_to_node,
|
|
metadata_dict_to_node,
|
|
node_to_metadata_dict,
|
|
)
|
|
|
|
|
|
def _to_metal_filters(standard_filters: MetadataFilters) -> list:
|
|
filters = []
|
|
for filter in standard_filters.legacy_filters():
|
|
filters.append(
|
|
{
|
|
"field": filter.key,
|
|
"value": filter.value,
|
|
}
|
|
)
|
|
return filters
|
|
|
|
|
|
class MetalVectorStore(VectorStore):
|
|
def __init__(
|
|
self,
|
|
api_key: str,
|
|
client_id: str,
|
|
index_id: str,
|
|
):
|
|
"""Init params."""
|
|
import_err_msg = (
|
|
"`metal_sdk` package not found, please run `pip install metal_sdk`"
|
|
)
|
|
try:
|
|
import metal_sdk # noqa
|
|
except ImportError:
|
|
raise ImportError(import_err_msg)
|
|
from metal_sdk.metal import Metal
|
|
|
|
self.api_key = api_key
|
|
self.client_id = client_id
|
|
self.index_id = index_id
|
|
|
|
self.metal_client = Metal(api_key, client_id, index_id)
|
|
self.stores_text = True
|
|
self.flat_metadata = False
|
|
self.is_embedding_query = True
|
|
|
|
def query(self, query: VectorStoreQuery, **kwargs: Any) -> VectorStoreQueryResult:
|
|
if query.filters is not None:
|
|
if "filters" in kwargs:
|
|
raise ValueError(
|
|
"Cannot specify filter via both query and kwargs. "
|
|
"Use kwargs only for metal specific items that are "
|
|
"not supported via the generic query interface."
|
|
)
|
|
filters = _to_metal_filters(query.filters)
|
|
else:
|
|
filters = kwargs.get("filters", {})
|
|
|
|
payload = {
|
|
"embedding": query.query_embedding, # Query Embedding
|
|
"filters": filters, # Metadata Filters
|
|
}
|
|
response = self.metal_client.search(payload, limit=query.similarity_top_k)
|
|
|
|
nodes = []
|
|
ids = []
|
|
similarities = []
|
|
|
|
for item in response["data"]:
|
|
text = item["text"]
|
|
id_ = item["id"]
|
|
|
|
# load additional Node data
|
|
try:
|
|
node = metadata_dict_to_node(item["metadata"])
|
|
node.text = text
|
|
except Exception:
|
|
# NOTE: deprecated legacy logic for backward compatibility
|
|
metadata, node_info, relationships = legacy_metadata_dict_to_node(
|
|
item["metadata"]
|
|
)
|
|
|
|
node = TextNode(
|
|
text=text,
|
|
id_=id_,
|
|
metadata=metadata,
|
|
start_char_idx=node_info.get("start", None),
|
|
end_char_idx=node_info.get("end", None),
|
|
relationships=relationships,
|
|
)
|
|
|
|
nodes.append(node)
|
|
ids.append(id_)
|
|
|
|
similarity_score = 1.0 - math.exp(-item["dist"])
|
|
similarities.append(similarity_score)
|
|
|
|
return VectorStoreQueryResult(nodes=nodes, similarities=similarities, ids=ids)
|
|
|
|
@property
|
|
def client(self) -> Any:
|
|
"""Return Metal client."""
|
|
return self.metal_client
|
|
|
|
def add(self, nodes: List[BaseNode], **add_kwargs: Any) -> List[str]:
|
|
"""Add nodes to index.
|
|
|
|
Args:
|
|
nodes: List[BaseNode]: list of nodes with embeddings.
|
|
|
|
"""
|
|
if not self.metal_client:
|
|
raise ValueError("metal_client not initialized")
|
|
|
|
ids = []
|
|
for node in nodes:
|
|
ids.append(node.node_id)
|
|
|
|
metadata = {}
|
|
metadata["text"] = node.get_content(metadata_mode=MetadataMode.NONE) or ""
|
|
|
|
additional_metadata = node_to_metadata_dict(
|
|
node, remove_text=True, flat_metadata=self.flat_metadata
|
|
)
|
|
metadata.update(additional_metadata)
|
|
|
|
payload = {
|
|
"embedding": node.get_embedding(),
|
|
"metadata": metadata,
|
|
"id": node.node_id,
|
|
}
|
|
|
|
self.metal_client.index(payload)
|
|
|
|
return ids
|
|
|
|
def delete(self, ref_doc_id: str, **delete_kwargs: Any) -> None:
|
|
"""
|
|
Delete nodes using with ref_doc_id.
|
|
|
|
Args:
|
|
ref_doc_id (str): The doc_id of the document to delete.
|
|
|
|
"""
|
|
if not self.metal_client:
|
|
raise ValueError("metal_client not initialized")
|
|
|
|
self.metal_client.deleteOne(ref_doc_id)
|