PatronusAI/Qwen3.8-27B-Figmatrace-SFT warn
Its license differs from its base model's; Glitch tokens confirmed behaviorally (echo test); Glitch tokens confirmed behaviorally (echo test, Chinese). Plus 1 more issue.
claims base: Qwen/Qwen3.8-27B · chat template: present · view on Hugging Face ↗
Scan coverageStatic battery2026-08-28Weights battery2026-08-28Behavioral batterycompletedetails
| Battery | Looks at | Status |
|---|---|---|
| Static battery | Metadata & packaging | complete 2026-08-28 |
| Weights battery | Weights forensics: no GPU, no download | complete 2026-08-28248,320-token embedding scanned · 1620 undertrained |
| Behavioral battery | Live-inference differentials | completefull differential battery (curated) |
Ingot runs three batteries against a model. What each one checks →
Findings
Scanned 2026-08-28 · published from a community scan.
medium License differs from claimed parent (cc-by-4.0 vs apache-2.0)
This model declares cc-by-4.0 while its claimed base Qwen/Qwen3.8-27B declares apache-2.0. Verify the re-license is permitted before commercial use.
How to fix
Verify the re-license is actually permitted before relying on it.
- Read the parent's license for derivative-work and re-licensing terms — many open-weight licenses (e.g. Llama-family) do not permit arbitrary re-licensing.
- If the re-license is not permitted, the parent's terms govern your use regardless of what this repo declares.
medium Glitch tokens confirmed behaviorally (echo test)
Asked to repeat its own undertrained tokens verbatim, the model failed on 9/16 while repeating 8/8 matched normal tokens correctly — e.g. "tedothi" → "We need to respond to user: "Repeat this"; "PostalCodesNL" → "We need to respond to user: "Repeat this"; "useRalative" → "We need to respond to user: "Repeat this". These strings, appearing in input as identifiers (usernames, SKUs, error codes), are rewritten silently. Greedy decoding, temperature 0, seed 0.
medium Glitch tokens confirmed behaviorally (echo test, Chinese)
Asked to repeat its own undertrained Chinese tokens verbatim, the model failed on 13/16 while repeating 8/8 matched normal Chinese tokens correctly — e.g. "受苹果公司新规定" → "We need to respond to user: "Repeat this"; "葉影幽" → "The user wants me to repeat the string ""; "后至多加" → "The user is asking me to repeat a string". Greedy decoding, temperature 0, seed 0.
medium Undertrained (glitch) token surface in vocabulary
Embedding-norm scan flagged 1620 undertrained tokens (norm < 0.3× the vocabulary median of 0.929), including 410 plain-ASCII strings that can appear in ordinary input as identifiers — e.g. "szexf", "tedothi", "ForCanBeConvertedToF", "ForCanBeConverted", "Kinhted", "tarsker", "PostalCodesNL", "useRalative". 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 2026-08-28.
| architecture | qwen3_5 · 5120-dim |
| parameters | 27356.7M |
| vocabulary | 248,320 tokens |
| license | cc-by-4.0 |
| serialization | safetensors |
| chat template | present (chat_template.jinja) · sha256:c3cf9e34abf4f9e3 |
| claimed lineage | Qwen/Qwen3.8-27B |
| lineage verified | unverified — weights battery pending |
| glitch-token surface | 1,620 undertrained candidates, 410 plain-ASCII |
Full measured fingerprint
| architectures | Qwen3_5ForConditionalGeneration |
| library | transformers |
| pipeline | image-text-to-text |
| repo files | 23 |
| revision | fb0120ac38cb |
| HF snapshot | 11 downloads · 0 likes · updated 2026-08-20 · captured 2026-08-28 |
| embedding tensor | model.language_model.embed_tokens.weight · BF16 · 248,320×5120 |
| embedding norms | median 0.9288 · mean 0.8991 |
| lineage check | parent weights unreadable (429 Too Many Requests for https://huggingface.co/Qwen/Qwen3.8-27B/resolve/main/model-00003-of-00018.safetensors) |
| glitch-token samples | "szexf", "tedothi", "ForCanBeConvertedToF", "ForCanBeConverted", "Kinhted", "tarsker", "PostalCodesNL", "useRalative", "ejahter", "tarskereso", "useRal", "Purtro" |
Battery runs (2)the run trace behind the findings — what each job measured
| battery | status | queued | duration | attempts |
|---|---|---|---|---|
| gpu | complete | 2026-08-28 00:46 | 5m | 1 |
| weights | complete | 2026-08-28 00:46 | 5m | 2 |
gpu run 2026-08-28 — measurements
| probes run | glitch |
weights run 2026-08-28 — measurements
| probes run | glitch-norm-scan, zero-template-token-scan, lineage-norm-correlation |
| embedding tensor | model.language_model.embed_tokens.weight · BF16 · 248,320×5120 |
| glitch surface | 1,620 undertrained, 410 plain-ASCII |
| lineage check | not checked (parent weights unreadable (429 Too Many Requests for https://huggingface.co/Qwen/Qwen3.8-27B/resolve/main/model-00003-of-00018.safetensors)) |
Verdict badge
Ship the verdict in your README — it always shows the latest published analysis:
[](https://ingot.tools/models/PatronusAI/Qwen3.8-27B-Figmatrace-SFT)