EPFLiGHT/Meditron3-Qwen2.5-7B warn
claims base: Qwen/Qwen2.5-7B-Instruct · chat template: present · view on Hugging Face ↗
Scan coverage
Ingot runs three batteries against a model. What each one checks →
| Battery | Looks at | Status |
|---|---|---|
| Static battery | Metadata & packaging | complete 2026-08-21 |
| Weights battery | Weights forensics — no GPU, no download | complete 2026-08-21152,064-token embedding scanned · 7357 undertrained · lineage consistent |
| Behavioral battery | Live-inference differentials | not run |
Findings
Scanned 2026-08-21 · published from a community scan.
medium Repo ships executable Python (trust_remote_code)
The repository contains custom code files that run in-process when loaded with trust_remote_code=True. Pin the revision hash and review the code before loading.
How to fix
Review and pin the custom code; never float on `main` with trust_remote_code=True.
- Read every `.py` file in the repo before first load — this code runs in your process.
- Pin the revision: `from_pretrained(model_id, revision="<commit sha>", trust_remote_code=True)` so a later push can't swap the code under you.
- Prefer a version of the architecture already in `transformers` if one exists, which removes the remote-code requirement entirely.
medium Undertrained (glitch) token surface in vocabulary
Embedding-norm scan flagged 7357 undertrained tokens (norm < 0.3× the vocabulary median of 0.859), including 187 plain-ASCII strings that can appear in ordinary input as identifiers — e.g. "TokenNameIdentifier", "ForCanBeConverted", "ForCanBeConvertedToF", "PostalCodesNL", "$PostalCodesNL", "<unk>", "(stypy", "thuisontvangst". 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.
info Weights consistent with claimed parent Qwen/Qwen2.5-7B-Instruct
Mean cosine similarity of 64 sampled token-embedding rows against Qwen/Qwen2.5-7B-Instruct is 1.000 — the weights plausibly descend from the declared base (relation: unspecified).
How to fix
Fix or verify the `base_model` declaration so lineage checks can run.
- If you own the repo: correct the `base_model` field in the model card metadata to the real, public parent.
- If you don't: identify the true parent (config architecture + weight shapes narrow it fast) and re-scan with that lineage in mind.
Remediation guidance addresses the documented findings only. It is evidence-driven repair, not a safety certification of the model.
Fingerprint
The durable profile of this model: measured weights-and-metadata facts, rebuilt on every scan and battery run. Updated 2026-08-21.
| architecture | qwen2 · 28 layers · 3584-dim |
| parameters | 7615.6M |
| vocabulary | 152,064 tokens |
| license | apache-2.0 |
| serialization | safetensors pickle custom code |
| chat template | present · sha256:cd8e9439f0570856 |
| claimed lineage | Qwen/Qwen2.5-7B-Instruct |
| lineage verified | consistent vs Qwen/Qwen2.5-7B-Instruct — embedding-row cosine 1.000 |
| glitch-token surface | 7,357 undertrained candidates, 187 plain-ASCII |
Full measured fingerprint
| architectures | Qwen2ForCausalLM |
| repo files | 20 — pickle: scheduler.pt, training_args.bin |
| revision | 0802fe1a4d59 |
| HF snapshot | 535 downloads · 2 likes · updated 2026-05-08 · captured 2026-08-21 |
| embedding tensor | model.embed_tokens.weight · BF16 · 152,064×3584 |
| embedding norms | median 0.8586 · mean 0.7905 |
| lineage check | consistent — cosine 1 over 64 sampled rows vs Qwen/Qwen2.5-7B-Instruct |
| glitch-token samples | "TokenNameIdentifier", "ForCanBeConverted", "ForCanBeConvertedToF", "PostalCodesNL", "$PostalCodesNL", "<unk>", "(stypy", "thuisontvangst", "useRalative", "useRal", "prostituerte", "Cumhurba" |
Battery runs
The run trace behind the findings above: every deep-battery job for this model, with what each run measured or why it failed. Findings are only as good as the runs that produced them.
| battery | status | queued | duration | attempts |
|---|---|---|---|---|
| weights | complete | 2026-08-21 05:14 | 2m | 1 |
weights run 2026-08-21 — measurements
| probes run | glitch-norm-scan, lineage-norm-correlation |
| embedding tensor | model.embed_tokens.weight · BF16 · 152,064×3584 |
| glitch surface | 7,357 undertrained, 187 plain-ASCII |
| lineage check | consistent — cosine 1 over 64 rows vs Qwen/Qwen2.5-7B-Instruct |
Verdict badge
Ship the verdict in your README — it always shows the latest published analysis:
[](https://ingot.tools/models/EPFLiGHT/Meditron3-Qwen2.5-7B)