weft-kernel 0.1.0__py3-none-any.whl

weft_kernel/payload/vector.py

@@ -0,0 +1,28 @@
+"""An embedding: an ordered, non-empty tuple of components.
+
+Settled in G4: persistence accepts a vector already computed elsewhere; it
+is never computed from content directly. `Node.embedding` is this type, or
+`None` before an embedding has been computed. Excluded from a node's
+content-addressed identity (`node.py`) because it is derived from content
+and would otherwise bind an id to whichever model produced it.
+"""
+
+from pydantic import BaseModel, ConfigDict, model_validator
+
+
+class Vector(BaseModel):
+    """An embedding: an ordered, non-empty tuple of components."""
+
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    values: tuple[float, ...]
+
+    @model_validator(mode="after")
+    def _non_empty(self) -> "Vector":
+        if not self.values:
+            raise ValueError("Vector must carry at least one dimension")
+        return self
+
+    @property
+    def dimension(self) -> int:
+        return len(self.values)