CALIBRATION-AWARE · REPRODUCIBLE · USER-OWNED

Quantize with
the signal intact.

Turn Hub models into validated GGUF artifacts. Spark 2.5 gets its own pinned runtime; every importance matrix is built, reopened, and proven before it can shape a quant.

✓ Immutable toolchains✓ Real matrix application✓ Final load probe

Notebook download needs no site sign-in. Private sources and personal Jobs do.

SOURCE

Choose the model

Inspection reads metadata only. Weights stay on the Hub until compute begins.

Model repository Revision

◇

Awaiting inspection

Architecture, format, exact revision, and size will appear here.

CALIBRATION

Teach the quant what matters

Add one or more text datasets. The worker normalizes, deduplicates, hashes, and preserves the combined corpus.

Build and apply importance matrix

Advanced matrix controls

Matrix mode Target chunks Minimum tokens Allow below floor

ARTIFACT

Name the result

Output ownership and visibility are independent of this public Space.

Quant type Output repository Visibility File basename optional Split limit Execution route Hardware Timeout

PREFLIGHT

Prove the route

No compute starts until architecture, matrix inputs, destination, and payer all agree.

Inspect the model, then run preflight.

COMPUTE PAYERNot signed inCenedril is never charged.

ACTIVE JOB

Submitted

Open Job ↗Open output ↗