Translates the single-model source into a concrete package manifest, compatibility key, lifecycle, and request contract for hyper-specialized tiny models.
Implementation evidence: this topic is grounded in the reviewed GGUF.MiRust.com source snapshot. It documents observed code and artifacts without claiming broad deployment, model quality, or production readiness.
Architecture topic: this page does not claim that the WordPress website implements or executes the described runtime behavior.
Package identity
Each skill needs a stable ID, semantic version, artifact SHA-256, source and license, model family, tokenizer ID and hash, format/version, quantization, task contract, authority scope, resource estimate, and evidence status.
Compatibility key
Independent models may differ in architecture and tokenizer. Shared-base adapters must match base artifact identity, target modules, rank/scaling convention, tokenizer, tensor names, and runtime adapter ABI.
Lifecycle
Installed, verified, resident, active, suspended, retired, and quarantined are separate states. The current load_model/free_model pair supplies only a single resident slot; a module registry needs handles and an eviction policy.
Request contract
Inputs and outputs require typed schemas, max sizes, confidence/uncertainty semantics, deterministic settings, cancellation, error codes, and side-effect authority. A role label alone is insufficient.
Scope
This starter page defines the questions, boundaries, evidence, and failure modes that should be recorded before a capability is presented as supported.
Engineering considerations
- Identify the source, version, target environment, and owner.
- Separate observed values from estimates and externally reported values.
- Record trade-offs, unsupported cases, and fallback behavior.
- Link performance statements to a compatible benchmark methodology.
Verification questions
- What exact artifact, revision, backend, and environment were reviewed?
- Which assumptions could change the result?
- Which data should be retained so another engineer can reproduce the conclusion?