Model page

numind/NuExtract3 warn

EOS ids disjoint between config.json and generation_config.json; the chat template differs from its base model, which changes behavior.

downloads 127.0klikes 344license apache-2.0arch qwen3_5params 4539.3Mupdated 2026-08-20

claims base: Qwen/Qwen3.5-4B · chat template: present · view on Hugging Face ↗

Scan coverageStatic battery2026-08-27Weights battery2026-08-27Behavioral batteryn/adetails
BatteryLooks atStatus
Static batteryMetadata & packagingcomplete 2026-08-27
Weights batteryWeights forensics: no GPU, no downloadcomplete 2026-08-27248,320-token embedding scanned · 0 undertrained · lineage consistent
Behavioral batteryLive-inference differentialsn/anot applicable: image-to-text model has no text-generation surface to probe

Ingot runs three batteries against a model. What each one checks →

Findings

Scanned 2026-08-27 · published from a community scan.

medium EOS ids disjoint between config.json and generation_config.json

config.json declares eos_token_id [248046] while generation_config.json declares [248044] with no overlap. Runtimes read one or the other, so at least one of them stops generation on the wrong token (or never). Align both files on the token the chat template actually ends turns with.

How to fixingot patch

Align the stop-token declarations — a pure metadata fix to `generation_config.json` (and `config.json`).

  1. Identify the token the chat template actually ends assistant turns with (e.g. `<|eot_id|>`, `<end_of_turn>`, `<|im_end|>`) and make sure its id is in `generation_config.json`'s `eos_token_id` list.
  2. Keep `config.json`'s `eos_token_id` consistent with (or a subset of) `generation_config.json`'s — runtimes differ in which file they read.
  3. For the pad-equals-EOS hazard: repoint `pad_token` at a dedicated padding token before fine-tuning; serving is unaffected.
  4. Until the repo is fixed, pass explicit stop tokens to your serving stack (e.g. vLLM `stop_token_ids`, llama.cpp `--override-kv tokenizer.ggml.eos_token_id`).

medium Chat template differs from claimed parent

The chat template does not match Qwen/Qwen3.5-4B's. Template drift silently changes model behavior even when weights are identical — 37% of drifted derivatives in our census left it undisclosed. Diff the templates before deploying.

How to fixingot patch

Restore the parent's chat template in `tokenizer_config.json` — a pure metadata fix.

  1. Run `ingot patch <owner/model>` — the patch manifest carries the parent's template and applies it to a local copy's `tokenizer_config.json`.
  2. Or fix by hand: copy the `chat_template` value from the parent repo's `tokenizer_config.json` into this model's, and pin your serving stack to that file.
  3. If the drift was intentional (the author retrained on a new template), confirm that in the model card before "fixing" it — restoring the parent template on retrained weights changes behavior too.

info Partial coverage — not a generative language model

image-to-text model — no token vocabulary, so chat-template, tokenizer, and behavioral checks are not applicable; packaging, license, and serialization (pickle) checks apply.

info Embedding-norm glitch scan clean

No undertrained tokens found: every non-special token's embedding norm is above 0.3× the vocabulary median (0.652). The glitch-token data-corruption class has no candidate surface in this model.

info Weights consistent with claimed parent Qwen/Qwen3.5-4B

Mean cosine similarity of 64 sampled token-embedding rows against Qwen/Qwen3.5-4B is 1.000 — the weights plausibly descend from the declared base (relation: unspecified).

Put this result in your workflow

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.

Fix it

Some findings are metadata-level and patchable — apply the fixes to your local copy (your weights never leave your machine):

npx @ingotai/scan patch numind/NuExtract3

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 2026-08-27.

architectureqwen3_5
parameters4539.3M
vocabulary248,320 tokens
licenseapache-2.0
serializationsafetensors
chat templatepresent (chat_template.jinja) · sha256:31e44d28615d268e
claimed lineageQwen/Qwen3.5-4B
lineage verifiedconsistent vs Qwen/Qwen3.5-4B — embedding-row cosine 1.000
glitch-token surfaceclean no undertrained tokens
Full measured fingerprint
architecturesQwen3_5ForConditionalGeneration
librarytransformers
pipelineimage-to-text
repo files29
revisionc99dc8f5641b
HF snapshot127.0k downloads · 344 likes · updated 2026-08-20 · captured 2026-08-27
embedding tensormodel.language_model.embed_tokens.weight · BF16 · 248,320×2560
embedding normsmedian 0.6518 · mean 0.6557
lineage checkconsistent — cosine 1 over 64 sampled rows vs Qwen/Qwen3.5-4B
Battery runs (1)the run trace behind the findings — what each job measured
batterystatusqueueddurationattempts
weightscomplete2026-08-27 22:2766s1
weights run 2026-08-27 measurements
probes runglitch-norm-scan, zero-template-token-scan, lineage-norm-correlation
embedding tensormodel.language_model.embed_tokens.weight · BF16 · 248,320×2560
glitch surface0 undertrained, 0 plain-ASCII
lineage checkconsistent — cosine 1 over 64 rows vs Qwen/Qwen3.5-4B

Verdict badge

Ship the verdict in your README — it always shows the latest published analysis:

Ingot verdict: warn

[![Ingot scan](https://ingot.tools/api/v1/models/numind/NuExtract3/badge.svg)](https://ingot.tools/models/numind/NuExtract3)
Gate it in CI