Operators

Operators describe what a model does without depending on model-family names or a specific runtime. Plans represent projections, normalization, rotary embeddings, attention, recurrent and convolutional state, nonlinear activation, residuals, selection, and input or output boundaries with explicit shapes and settings.

A compiler profile lists the operators it can execute. Unknown or unsupported operations appear in PlanCoverage and stop compilation. PLLM never silently replaces them with plaintext or approximate work. Research methods also use these shared operators instead of relying on model-specific weight names.

An operator in a model plan shows only that PLLM can describe it. Read numeric semantics and model quality for why operator coverage alone is insufficient. See compiler for execution support and numerics for integer and approximation requirements.

The bounded SiLU activation is the first operator with an experimental plan-bound protected executor. Other listed operators remain semantic records until their compiler and runtime coverage is implemented.

Python SDK example

from pathlib import Path

from pllm import lower_model

fixture = Path("crates/pllm-models/tests/fixtures/mini-coder-4b-c87892d-config.json")
plan = lower_model(fixture.read_bytes(), batch=1, max_input_tokens=16, max_new_tokens=4)
operators = {operation["operator"] for operation in plan.prefill["operations"]}
assert {"linear", "rotary_embedding", "silu", "softmax"} <= operators
assert plan.coverage().complete is False

API: Python SDK objects and signatures

Run this example from a PLLM source checkout; package installations do not include the checked-in fixture. This inspects semantic operators; it does not establish execution coverage.