meta-llama/Prompt-Guard-86M warn
Glitch tokens that can silently corrupt ordinary input.
chat template: not found · view on Hugging Face ↗
Scan coverageStatic battery2026-08-20Weights battery2026-08-20Behavioral batteryn/adetails
| Battery | Looks at | Status |
|---|---|---|
| Static battery | Metadata & packaging | complete 2026-08-20 |
| Weights battery | Weights forensics: no GPU, no download | complete 2026-08-20251,000-token embedding scanned · 4559 undertrained |
| Behavioral battery | Live-inference differentials | n/anot applicable: text-classification model has no text-generation surface to probe |
Ingot runs three batteries against a model. What each one checks →
Findings
Scanned 2026-08-20 · published from a community scan.
info Gated repository
Access requires accepting the owner's terms; check the gate conditions for redistribution and field-of-use limits.
How to fix
Read the gate terms before building on the model.
- Check the gate conditions on the Hugging Face repo for redistribution and field-of-use limits — they bind your deployment, not just your download.
medium Undertrained (glitch) token surface in vocabulary
Embedding-norm scan flagged 4559 undertrained tokens (norm < 0.3× the vocabulary median of 4.049), including 123 plain-ASCII strings that can appear in ordinary input as identifiers — e.g. "<0xFD>", "<extra_id_26>", "<extra_id_48>", "<extra_id_97>", "<extra_id_69>", "<extra_id_92>", "<extra_id_21>", "<extra_id_43>". In models where this class was tested behaviorally, such tokens silently rewrote user input into confident, schema-valid, wrong output. These are candidates from the weights alone; behavioral confirmation requires the behavioral battery.
How to fixruntime guardweight-level
Keep the affected token strings out of the model's input — the scan-derived runtime guard carries this model's exact blocklist.
- Fetch this model's guard artifact (`/api/v1/guard/<owner>/<model>`): the confirmed corrupting tokens and the low-norm candidate list, derived from the published scan.
- Screen inbound text with it (the `@ingotai/guard` package is a reference implementation) and route flagged records to a different model or human review — verbatim-copy tasks on flagged strings are the failure mode.
- The underlying cause is undertrained embeddings in the weights; a true fix is weight-level (continued pretraining on the affected tokens) — that is not a patch, it's a training job.
Remediation guidance addresses the documented findings only. It is evidence-driven repair, not a safety certification of the model.
Check every checkpoint before it ships
Use the web app, API, CLI, or CI gate to scan candidate checkpoints and catch model drift before deployment. Public-model scans publish to the open database; paid plans add the volume needed for continuous checks.
Fingerprint
The durable profile of this model: measured weights-and-metadata facts, rebuilt on every scan and battery run. Updated 2026-08-20.
| architecture | deberta-v2 |
| parameters | 278.8M |
| vocabulary | 251,000 tokens |
| license | llama3.1 |
| serialization | safetensors |
| chat template | none |
| glitch-token surface | 4,559 undertrained candidates, 123 plain-ASCII |
Full measured fingerprint
| architectures | DebertaV2ForSequenceClassification |
| library | transformers |
| pipeline | text-classification |
| repo files | 11 |
| gated | manual |
| revision | 1209add6ca7d |
| HF snapshot | 4.5M downloads · 393 likes · updated 2025-11-12 · captured 2026-08-20 |
| embedding tensor | deberta.embeddings.word_embeddings.weight · F32 · 251,000×768 |
| embedding norms | median 4.0491 · mean 3.9669 |
| lineage check | no claimed base model |
| glitch-token samples | "<0xFD>", "<extra_id_26>", "<extra_id_48>", "<extra_id_97>", "<extra_id_69>", "<extra_id_92>", "<extra_id_21>", "<extra_id_43>", "<extra_id_29>", "<extra_id_27>", "<extra_id_75>", "<extra_id_37>" |
Battery runs (1)the run trace behind the findings — what each job measured
| battery | status | queued | duration | attempts |
|---|---|---|---|---|
| weights | complete | 2026-08-20 18:02 | 31s | 1 |
Verdict badge
Ship the verdict in your README — it always shows the latest published analysis:
[](https://ingot.tools/models/meta-llama/Prompt-Guard-86M)