Skip to content

v0.20.0 — observed Inspect GGUF provider

Choose a tag to compare

@Sahil170595 Sahil170595 released this 12 Oct 14:12
f322f30

Quantfit 0.20.0 releases PR #123 individually: observed Inspect GGUF provider.

Released independently from PR #123: observed Inspect GGUF provider.

Original introduction notes follow; the linked completed source acceptance supersedes their historical pending-qualification wording.

  • Optional public Inspect GGUF provider/CLI observations with actual weight/binary
    hashes, immutable Hub pins, applied CPU controls and owned cancellable server
    lifecycle. Managed paired runs require80native calls/full40probes before one
    pinned judge batch, close both resident models before judging and enforce a
    conservative RAM estimate. Local synthetic evidence and pending hosted installed
    model qualification: validation/2026-10-09-inspect-gguf/; no safety/reproduction GO.

The introduced source passed hosted acceptance; release publication separately qualifies the exact versioned distributions. Historical producer versions and run records remain unchanged. No QSR v1 freeze or new scientific GO is claimed.

Local validation: 1774 sparse local tests passed with 32 skipped, plus two separately available numerical properties; full Torch/model proof is hosted. Exact-head hosted CI passed before merging. The protected publishing workflow repeats full qualification and attests the exact distributions before PyPI publication.