GGUF tensor inventory

Inspect the tensor assignment, not only the filename's quantization label.

A GGUF artifact can mix Q4, Q5, Q6, integer, and floating-point encodings. The useful evidence is the tensor-by-tensor assignment and its reproducible digest.

Run a bounded inventory

npx -y deepbom@1.97.4 gguf "./model.gguf" --tensors
npx -y deepbom@1.97.4 gguf "./model.gguf" --tensors --compact \
  --tensor-offset 0 --tensor-limit 100

The table reports tensor name, serialized encoding, shape in GGUF ne0-first order, effective bits per element, and byte range. The assignment digest is computed from a documented normalized projection so another implementation can reproduce it.

Evidence limits

File evidence	Serialized tensor directory, data offsets, metadata, payload bounds, and exact file hash.
Deterministic derivation	Encoding histogram, effective storage bits, and normalized tensor-assignment digest.
External evidence	Base-model lineage, calibration data, importance-matrix provenance, QAT or PTQ history, task accuracy, and runtime placement.

This guide deliberately does not redistribute or claim measurements for a third-party GGUF. Bind any published result to an immutable source revision and SHA-256.