ASSERT

Does this pack keep what it claims?

Quantization can drop or disable part of a model without failing. The pack loads, serves, and answers correctly — while a component its card says it preserved is either gone, or present and ignored by the runtime. Nothing errors. Nothing in the logs says so.

This reads a published repo's metadata only: safetensors headers and config.json, fetched over ranged requests, from your browser. No weights are downloaded, so a 27 GB pack costs a few kilobytes to check.

Our packs: · · ·
Elsewhere: ·

What it checks

  1. Does every exclusion entry name a real module? An entry naming something this architecture does not have protects nothing, and whatever it was meant to keep was quantized. The common case is a vision tower: Llama and Pixtral call it vision_tower, Qwen calls it visual, and the documented default names the former.
  2. Did any module of the source model fail to reach the pack? from_pretrained does not materialise tensors the AutoModel class has no slot for, so an auxiliary head living inside the checkpoint never reaches the quantizer and never reaches the artifact.
  3. Is anything at source precision without being declared? A runtime builds those quantized, looks for a packed weight, finds a plain one, and skips it — leaving the module randomly initialised.

The first two are observations. The third is a prediction, and says so where it appears.

Why the rules are per format

Every format declares exclusions differently, and reading one convention into another is how a checker invents verdicts. A bare mtp covers a whole head under AWQ's substring rule and covers nothing under compressed-tensors' exact-match rule.

FormatFieldHow an entry matches
compressed-tensorsignoreexact equality, or a re: regex
awqmodules_to_not_convertplain substring
gptqdynamicordered, first match wins; -: excludes
bitsandbytesllm_int8_skip_modulesexact path component, or a cumulative prefix
modeloptexclude_modulesexact, substring, or an fnmatch glob
auto-roundextra_configper-module bits ≥ 16, or a block allowlist

Every regex above resolves through Python's re.match, which anchors at the start: re:layers.0 does not match mtp.layers.0…. These were read out of vLLM's source, not from documentation. A format not on the list is reported as not checked rather than assumed healthy — a checker that cannot tell "healthy" from "not checked" is not a checker.