SmolLM2-135M
The smallest SmolLM2 base model — a compact 135M-parameter Llama-architecture LM for on-device text generation and as a fine-tuning base.
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
The uploader states: "SmolLM2 is a family of compact language models available in three size: 135M, 360M, and 1.7B parameters. They are capable of solving a wide range of tasks while being lightweight enough to run on-device." This 135M entry is the base pre-trained model, trained on ~2 trillion tokens (FineWeb-Edu, DCLM, The Stack, plus new filtered datasets). The card notes the models "primarily understand and generate content in English" and that generated content "may not always be factually accurate, logically consistent, or free from biases." HF tags: text-generation, transformers, safetensors, en, llama, text-generation-inference (no `conversational` tag).
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.
“This is a base / pretrained language model (the SmolLM2 family base rung), not instruction-tuned.”
Ardora reads this as a plain text-continuer with no assistant/chat register (assistant-adherence band 1, high confidence; ardora_score band 1). That is exactly the disposition of an untuned base, so the uploader's "base / pretrained" identity is supported by the reading.
“A compact model, lightweight enough to run on-device (the 135M rung of the sweep).”
Provenance-corroborated: the parameter count and architecture were verified from the model's config.json and recorded in the atlas provenance. This is a catalogued fact, not a disposition read.
“"Capable of solving a wide range of tasks."”
"Solving a wide range of tasks" is a task-competence claim. Ardora reads disposition (what the model tends to do), not capability, so it neither confirms nor scores this.
— nothing demonstrated to witness; this claim is out of Ardora’s disposition scope.
“General-purpose; the card names no specialized domain.”
The reading finds no pull toward any specialized domain idiom (domain-specialization band 1); consistent with the card's general-purpose framing.
“The model "primarily understand[s] and generate[s] content in English."”
Ardora's reading is English-centric and explicitly abstains on multilingual disposition; it can neither confirm nor contradict the "primarily English" claim at the published ceiling.
— nothing demonstrated to witness; this claim is out of Ardora’s disposition scope.
“Pretrained on ~2 trillion tokens (FineWeb-Edu, DCLM, The Stack, plus filtered datasets).”
Training-token counts and corpus composition are provenance the atlas records best-effort from the card; Ardora does not independently verify training data, so this is out of the reading's scope.
— nothing demonstrated to witness; this claim is out of Ardora’s disposition scope.
“Improves over SmolLM1 across knowledge / reasoning / instruction benchmarks (HellaSwag, ARC, MMLU, PIQA...).”
Benchmark scores and cross-version capability comparisons are outside Ardora's disposition scope. The reading confirms the base's plain-continuer disposition but does not score capability.
— 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/pretrained model for on-device text generation and downstream fine-tuning.
training-data notes
Pretrained on ~2T tokens: FineWeb-Edu, DCLM, The Stack, plus curated math/code (SmolLM2 model card & paper). Arch verified from config.json: hidden_size 576, 30 layers, 9 heads (3 KV), vocab 49152.
lineage
A base model — the root of its lineage.
among its kin — the matchups
Compare side by side →Its instruction-tuned finetune follows instructions and holds an assistant register; this base does neither — it continues text plainly. The shift is clear and high-confidence.
larger siblingSmolLM2-360MBasescoreminimalThe same plain-continuer role one size up in the SmolLM2 sweep (360M vs 135M); both read as bases with no assistant register.
larger siblingSmolLM2-1.7BBasescoreminimalThe 1.7B base at the top of the SmolLM2 sweep — same plain-continuer role, larger capacity.