Model-load failure runbook

Provides an operator sequence for distinguishing network failure, transfer failure, parser rejection, tensor rejection, scratch/cache allocation failure, and provenance failure.

Experimental
Last verified
2026-06-25 00:00 UTC
Updated
Reading time
2 minutes

Provides an operator sequence for distinguishing network failure, transfer failure, parser rejection, tensor rejection, scratch/cache allocation failure, and provenance failure.

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.

1. Classify the stage

UI state Likely stage Evidence
fetch error HTTP route or response Status, route, server log
WASM allocation failed Zero/over-limit or memory pressure Artifact bytes, 128 MiB cap, console
Model rejected SLM parser/model admission Error code, result, diagnostics
Manifest error only Sidecar fetch/parse Manifest route and text

2. Preserve identity

Record model path, byte count, SHA-256, internal SLM checksum, WASM SHA-256, browser, source revision, and UTC time before retrying.

3. Reproduce natively

Run packer validation and manifest validation against the same bytes. A browser-only failure then belongs to transfer, memory, or host integration rather than the file schema alone.

4. Recover

Select another known artifact or reload corrected bytes. Do not relabel a rejected artifact as compatible by changing only the sidecar.

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?