Skip to main content

clika_runtime.runtime functions

from_graph​

from_graph(*args, **kwargs)

from_graph(graph: clika_runtime._core.io.ModelGraph, *, kv_batch: int = 1, kv_past_len: int = 0) -> clika_runtime._core.runtime.GraphNode

A serving node from an executable graph. CONSUMES graph (the node owns it; the ModelGraph object is spent afterwards). kv_past_len sizes a KV-stateful graph's per-session cache planes; a stateless graph ignores both knobs.

pipeline​

pipeline(*args, **kwargs)

pipeline(steps: list, *, external_inputs: collections.abc.Sequence[clika_runtime._core.TensorSpec] = [], external_outputs: collections.abc.Sequence[clika_runtime._core.TensorSpec] = []) -> clika_runtime._core.runtime.Pipeline

Build a pipeline from step dicts (keys: name, node, and optionally exec, input_map, output_map) plus the external I/O schema (TensorSpec lists: the request inputs the pipeline accepts and the response outputs it emits). For a single step with no input_map/output_map, omitted external specs default to that step's own schema; every other topology states them explicitly. The pipeline holds its nodes alive.