westlake-repl/Evolla-10B-hf warn
Its tokenizer differs from its claimed base model; glitch tokens that can silently corrupt ordinary input; the weights don't match the model it claims to be based on.
Could not load this model from the Hugging Face API (private, gated, or nonexistent). Findings below are from our archive.
Scan coverageStatic battery2026-08-21Weights battery2026-08-21Behavioral batterynot rundetails
| Battery | Looks at | Status |
|---|---|---|
| Static battery | Metadata & packaging | complete 2026-08-21 |
| Weights battery | Weights forensics — no GPU, no download | complete 2026-08-21128,256-token embedding scanned · 461 undertrained · lineage inconsistent |
| Behavioral battery | Live-inference differentials | not run |
Ingot runs three batteries against a model. What each one checks →
Findings
Scanned 2026-08-21 · published from a community scan.
medium Vocabulary size differs from claimed parent (128256 vs 446)
A changed vocab means changed tokenization: strings will split differently than on westlake-repl/SaProt_650M_AF2, which can shift behavior on identifiers, codes, and non-English text.
How to fixweight-level
Not patchable: the vocab size mirrors the embedding matrix in the weights. Verify the change was intentional.
- Do not edit `vocab_size` in config.json to "match the parent" — it must equal the embedding table in the shipped weights or the model won't load.
- Diff the tokenizers (`tokenizer.json` / added_tokens) against the parent to see what was added or removed, and test your own identifiers, codes, and non-English text through both.
- If the drift is unexplained by the model card, treat tokenization-sensitive behavior as unvalidated on this model.
medium Undertrained (glitch) token surface in vocabulary
Embedding-norm scan flagged 461 undertrained tokens (norm < 0.3× the vocabulary median of 0.601), including 131 plain-ASCII strings that can appear in ordinary input as identifiers — e.g. "$PostalCodesNL", "ForCanBeConverted", "TokenNameIdentifier", "useRalative", "ForCanBeConvertedToF", "PostalCodesNL", "ilmektedir", "CLIIIK". 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.
medium Weights inconsistent with claimed parent westlake-repl/SaProt_650M_AF2
This model declares westlake-repl/SaProt_650M_AF2 as its base (relation: unspecified), but its token-embedding geometry is incompatible: 4096-dim embeddings vs the parent's 1280-dim. A finetune cannot change embedding width — the lineage label is wrong or misleading. Treat provenance claims on this repo (training data, safety posture, licensing) as unverified.
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 | EvollaModel · 32 layers · 4096-dim |
| parameters | 10392.1M |
| vocabulary | 128,256 tokens |
| license | mit |
| serialization | safetensors pickle |
| chat template | present · sha256:ba03a121d097859c |
| claimed lineage | westlake-repl/SaProt_650M_AF2, meta-llama/Llama-3.1-8B-Instruct |
| lineage verified | inconsistent vs westlake-repl/SaProt_650M_AF2 |
| glitch-token surface | 461 undertrained candidates, 131 plain-ASCII |
Full measured fingerprint
| architectures | EvollaForProteinText2Text |
| repo files | 32 — pickle: pytorch_model-00001-of-00009.bin, pytorch_model-00002-of-00009.bin, pytorch_model-00003-of-00009.bin, pytorch_model-00004-of-00009.bin, pytorch_model-00005-of-00009.bin, pytorch_model-00006-of-00009.bin, pytorch_model-00007-of-00009.bin, pytorch_model-00008-of-00009.bin, pytorch_model-00009-of-00009.bin |
| revision | ca9a046b9106 |
| HF snapshot | 73.9k downloads · 1 likes · updated 2026-04-08 · captured 2026-08-21 |
| embedding tensor | model.embed_tokens.weight · F32 · 128,256×4096 |
| embedding norms | median 0.6012 · mean 0.5911 |
| lineage check | inconsistent — cosine undefined over undefined sampled rows vs westlake-repl/SaProt_650M_AF2 |
| glitch-token samples | "$PostalCodesNL", "ForCanBeConverted", "TokenNameIdentifier", "useRalative", "ForCanBeConvertedToF", "PostalCodesNL", "ilmektedir", "CLIIIK", "_ComCallableWrapper", "krvldkf", "webElementXpaths", "useRalativeImagePath" |
Battery runs (1)the run trace behind the findings — what each job measured
| battery | status | queued | duration | attempts |
|---|---|---|---|---|
| weights | complete | 2026-08-21 05:09 | 82s | 1 |
weights run 2026-08-21 — measurements
| probes run | glitch-norm-scan, lineage-norm-correlation |
| embedding tensor | model.embed_tokens.weight · F32 · 128,256×4096 |
| glitch surface | 461 undertrained, 131 plain-ASCII |
| lineage check | inconsistent — cosine undefined over undefined rows vs westlake-repl/SaProt_650M_AF2 |
Verdict badge
Ship the verdict in your README — it always shows the latest published analysis:
[](https://ingot.tools/models/westlake-repl/Evolla-10B-hf)