VulcoraTalk to us
← All scansHugging Face model
huihui-ai

Qwen2.5-0.5B-Instruct-abliterated

SuspiciousSomething is off — read before you trust it.
#guardrails-stripped
7,381 downloadsassessed Jul 2026read offline · never run
Book the real read →
Also onHuggingFaceArdora atlas

A real read — read offline against the base, never run. The verdict is reproducible.

◆ Vulcora caught it — offline, against its base, with proof

Vulcora reads the weights
Off-the-shelf tools · run on this model0 of 1 that could read it flagged it
Capability — not run on this modelwhat these tools read and what question they answer — not a result
  • model-signingOpenSSF / Sigstoreprovenance

    reads Sigstore signature bundles

    verifying a cryptographic provenance signature over model files; it can only attest what a publisher signed.

    requires a published Sigstore bundle; no scanned model ships one

1 off-the-shelf tool could read this model’s serialized files, and none flagged it. None of them reads what the weights mean: they hunt load-time code-execution, repo smells, or a provenance signature — real jobs, none of which is weight-level tampering. 3 other tools had nothing of the kind they read here. Vulcora reads the weights — offline, and never runs the model.

Couldn't reach the attestation service to re-verify just now.

Deeper readings

the full disposition & integrity reads — worth craving
Ardora · dispositionRead the full reading
minimalslightmoderatepronounceddominantAssistant Adherence: pronouncedRegister / Formality: abstained — not characterised at the published ceilingReasoning Scaffolding: slightDomain Specialization: minimalVerbosity: abstained — not characterised at the published ceilingTurn-Taking / Interactivity: moderatePerformed Voice: minimalAssistant○ RegisterReasoningDomain○ VerbosityTurn-TakingVoiceAssistant AdherenceRegister / FormalityReasoning ScaffoldingDomain SpecializationVerbosityTurn-Taking / InteractivityPerformed Voice

rings, inner → outer: minimal · slight · moderate · pronounced · dominant · ○ = abstained

The coarse per-axis disposition — abstains shown as honest gaps. Read the full reading for the character & anchors.

Read in fullArdora reading

Details

from the artifact’s public metadata
Architecture
Qwen2
Base
Qwen/Qwen2.5-0.5B-Instruct · Qwen
Downloads
7.4k
License
apache-2.0
Parameters
0.5B

Think this verdict is wrong?

Signed-in users can formally refute a read — Vulcora re-reads the model against your claim and records the outcome. You'll need an account to file one.

Sign in to refute ↗

Refutations are actor-gated — a guest can't file one.

Cited in a research corpus

someone else’s label — not Vulcora’s live verdict

This model appears as a labelled specimen in a public benchmark corpus. Each panel below states the benchmark authors’ own published ground-truth label and, where the research harness actually ran, its coarse readout — distinct from the Vulcora verdict above.

  • HuggingFace publisher-declared modifications specimen · huihui-ai/qwen2.5-0.5b-instruct-abliterated
    Benchmark’s published labelpoisonedattack family: safety_removal
    Research-harness readcited label · harness not yet run

    Published-benchmark / research-harness record (benchmark_import). Not a Vulcora live scan — it is the corpus’s statement about this model, shown here for provenance.

Want a verdict you can prove — on your own model?

Book the real read →Look up another →