For AI agents: the complete documentation index is available at https://a3s-lab.github.io/Power/v0.9.0/en/llms.txt, the full documentation bundle is available at https://a3s-lab.github.io/Power/v0.9.0/en/llms-full.txt, and this page is available as Markdown at https://a3s-lab.github.io/Power/v0.9.0/en/architecture.md.
  • English
  • v0.9.0
  • Embedded Inference Architecture

    Power treats inference as a bounded operation whose identity and effects can be verified. It does not treat a model name as an execution contract.

    Derive the boundary from first principles

    ConstraintRequired contractRuntime mechanism
    Memory, compute, and queue capacity are finite.Every request needs explicit limits and cancellation.Bounded admission, deterministic microbatching, placement plans, and session pools.
    A filename or model alias does not identify executed bytes.Artifact identity must survive storage, mirrors, and device placement.SHA-256 descriptors, signed identities, verified mirrors, and residency evidence.
    A response alone does not reveal the path that produced it.Runtime policy, device path, input, and output must be committed together.Canonical execution receipts and accelerator evidence.
    The service cannot be trusted to approve its own claims.Acceptance policy belongs outside the execution boundary.Nonce-bound TEE evidence and an independent verifier.

    One core, three surfaces

    revision-locked bundle       model-owned graph       API client
              |                         |                     |
        provisioner                 embedded               service
              \                         |                    /
               +----------- shared runtime core -----------+
                                      |
                  admission / placement / cancellation
                                      |
                        devices / weights / state
                                      |
                       evidence / canonical receipt
                                      |
                          independent verification

    The embedded library and hosted service are entry points into the same contracts, not two model implementations. Artifact provisioning is independent because downloading and installing a reviewed bundle is a different authority from executing it.

    Ownership is explicit

    Power ownsModel-owning crates own
    Typed CPU, CUDA, and Metal devices; bounded graph executionArchitecture, topology, layers, kernels, and arithmetic
    Admission, session pools, microbatching, cancellation, and limitsTokenizer, preprocessing, postprocessing, and generation policy
    Artifact identity, replicas, mirrors, placement, and residencyRevision pins, conversion, tensor contracts, and quality gates
    TEE privacy, attestation binding, sealed state, and receiptsKV/recurrent layout and semantic state

    This boundary keeps Power model-neutral. Language, vision, OCR, embedding, and future model crates can share resource and evidence machinery without moving their semantics into a central model switch statement.

    The runtime contracts

    Bounded execution

    Admission limits active and queued work. Device admission prevents independent models from overcommitting the same accelerator. Cancellation is checked before admission and remains safe while a request waits or executes.

    Verified weights

    Weight descriptors bind tensor ranges to storage identities. Complete and partial mirrors retain that identity across storage tiers. Placement and residency evidence record the actual selected path rather than the preferred path alone.

    Accelerator evidence

    Execution receipts can include the concrete device, fallback, fused-batch, or multi-device mesh selected for a declaration-bound execution. Evidence must match the model, runtime device, input digest, and output digest before it can be attached to a receipt.

    Recoverable state

    Authenticated sealed-state envelopes bind model and runtime identity to warm state. Recovery policy distinguishes primary and backup sources, authorizes export scopes, detects rollback, and zeroizes sensitive material.

    Private observability

    Digest-only receipts and telemetry make execution inspectable without requiring prompt or response content. Opaque renderer paths omit claims they cannot derive instead of fabricating deterministic evidence.

    Execution lifecycle

    declare limits and identity
              |
              v
    admit model + device capacity
              |
              v
    resolve storage + placement
              |
              v
    execute reviewed model plan
              |
              v
    commit input/output/device evidence
              |
              v
    release permits and verify receipt

    The detailed design includes tensor batches, residency budgets, partial mirrors, prefetch hints, heterogeneous meshes, sealed state, and tuning evidence. Read the canonical architecture document for the complete APIs, invariants, and validation gates.