TinyMistral-248M
A 248M from-scratch Mistral-architecture LM by Locutusque, showing small models can pretrain cheaply; a base for fine-tuning.
The character read stayed in shadow — the read did not separate one character clearly at the published ceiling. Published as a blank, not a guess.
the Silhouette
its disposition, drawn — the model's identity as a shape.
rings, inner → outer: minimal · slight · moderate · pronounced · dominant · ○ = abstained
minimal - a plain continuer with little disposition to characterize (clearly read)
disposition-definition (descriptive): how pronounced and clearly-read the model's overall character is, discounted by what had to be abstained. NOT a quality, capability, or safety ranking. A plain base reads low because it has less disposition to characterize, never because it is worse.
A base model — the champion's raw kit.
There is no finetune to read here. What it tends to do on its own is the Silhouette above; a build's change is read on its instruct sibling.
the disposition line
The Silhouette above, read as a line. Toggle to its per-trait readings; the abstains stay in plain sight below.
Coarse public bands generalized from the reading, descriptive and coverage-bounded; measured against the published fidelity ceiling for this engine version. Not the engine's raw numbers.
Characterization is descriptive and coverage-bounded; preview engine, single-battery read; see published fidelity ceiling.
what stayed in shadow — the abstains
the full record — the card answered · provenance · lineage
the uploader's card, answered
the uploader’s model card · as read 2026-07-07
Locutusque/TinyMistral-248M
A pre-trained language model, based on the Mistral 7B model, has been scaled down to approximately 248 million parameters. This model has been trained on 7,488,000 examples. This model isn't intended for direct use but for fine-tuning on a downstream task. ... trained on the Skylion007/openwebtext and JeanKaddour/minipile datasets ... pretrained on a single GPU (Titan V).
sourcehuggingface.co/Locutusque/TinyMistral-248M ↗
Verbatim from the public HuggingFace card as of 2026-07-07. A ~248M from-scratch Mistral-ARCHITECTURE base LM (a downscaled Mistral, pretrained from scratch on ~7.49M examples — not weight-derived from Mistral 7B). Arch verified from config.json (Mistral, hidden 1024, 12 layers, vocab 32005). Base rung of the TinyMistral pair.
the card, answered — claim by claim
Each row is a claim the uploader makes on their public card, quoted and attributed to them. Beside it is Ardora’s stance — coarse, and traced to a replayable witness or an honest abstain. Ardora reads what this model is; a capability, benchmark, safety, or language claim is out of scope and abstained, never refuted.
“the card describes this as 'a pre-trained language model' — a base model”
The disposition reads as a plain text-continuer with no assistant register (assistant-adherence reads minimal) — exactly what a pretrained base should read as. Witnessed as the TinyMistral-248M base reading.
“the card states it 'isn't intended for direct use but for fine-tuning on a downstream task'”
Consistent with the base reading: a plain continuer with no assistant/turn-taking register — reads as a base for fine-tuning, not a direct-use assistant. Witnessed as the TinyMistral-248M base reading.
“the card states it is 'based on the Mistral 7B model, ... scaled down to approximately 248 million parameters' (a downscaled Mistral)”
Provenance / architecture fact-check: config.json confirms the Mistral architecture family (hidden 1024, 12 layers, vocab 32005) recorded in provenance. A provenance verification, not a disposition stance.
disposition onlyArchitecture-family fact-check. Per provenance this is a from-scratch downscaled Mistral-ARCHITECTURE model, not weight-derived from Mistral 7B; the check confirms the architecture family, not a lineage of weights.
— a provenance fact-check, not a witnessed disposition reading.
“the card states 'approximately 248 million parameters'”
Provenance fact-check: config cross-check confirms ~248M parameters recorded in provenance. A provenance verification, not a disposition stance.
— a provenance fact-check, not a witnessed disposition reading.
“the card states it was 'trained on 7,488,000 examples' from the openwebtext and minipile datasets”
A training-corpus / provenance claim (dataset and example count). Not a disposition Ardora reads; out of scope. The disposition reading confirms only that it reads as a plain continuer.
disposition onlyTraining-data provenance claim — out of disposition scope; abstained, not refuted.
— nothing demonstrated to witness; this claim is out of Ardora’s disposition scope.
Ardora's reading of the HF data as of 2026-07-07.
The claims above are the uploader’s, quoted from their public model card as of 2026-07-07; the stances are Ardora’s, each traced to a replayable witness or an honest abstain. Ardora reads what this model is — not whether it is safe.
○ Claims are the uploader's, quoted from their public card at the capture date; stances are coarse, witnessed, disposition-only, measured against the published fidelity ceiling for this engine version -- not the Engine's raw numbers. Recomputed when the ceiling moves; capability, benchmark, and safety claims are abstained, never refuted.
the witness
analyzed 2026-07-05 · engine ardora-core-preview
provenance
Base model intended primarily for downstream fine-tuning and small-model research.
training-data notes
Pretrained from scratch (~7.49M examples from Skylion007/openwebtext + JeanKaddour/minipile) on a single GPU; a downscaled Mistral. Release date approximated from config transformers_version 4.35.0. Arch verified from config.json: Mistral, hidden 1024, 12 layers, 32 heads (8 KV), sliding_window 32, vocab 32005.
lineage
A base model — the root of its lineage.