huggingfacetb/smollm2-135m-instruct

Provenance & integrity

Revisionsha:12fd25f7…380bac

Weight fingerprintsha256:5af571cb…68ab8c

config.json oidoid:36293b60…6d80d0

Parameters135M· 134,515,008 params

Licenceapache-2.0

Released2024-11

MakerHugging Face (HuggingFaceTB)

Last modified2025-09-22

Accessopen access

Identity & integrity of the public artifact captured 2025-09-22 — the same file, unchanged, that Ardora read. Not a safety statement.

The reading

Character abstainedthe read did not separate one character clearly at the published ceiling

Fine-tune change · clear · high confidence

Instruction tuning (SFT + DPO) added a clear assistant disposition where the base is a plain continuer. The shift reads clearly and consistently.

Disposition

instruction-followinghigh confidence

markedly more than its base

assistant/chat registerhigh confidence

clearly present; the base is a plain text-continuer

Axis profile

Overall · band 3 · high confidence

moderate - a clearly-read but modestly-defined turn-taking assistant character

Assistant Adherencepronounced - reliably takes the assistant turn and follows the instruction shape; the base is a plain continuerhigh

Register / Formality— not characterised · the read battery does not calibrate tone / diction; register-formality is not characterized at the published ceiling

Reasoning Scaffoldingslight - mostly answers directly; no staged deliberation readlow

Domain Specializationminimal - general-purpose; no domain-idiom pull (domain EXPERTISE is a separate, abstained capability question)medium

Verbosity— not characterised · default response length / elaboration is not exercised by the read battery; not characterized at the published ceiling

Turn-Taking / Interactivitymoderate - takes an assistant turn; follow-up / clarify behavior not separately exercisedlow

Performed Voiceminimal - neutral assistant / tool voice; no sustained charactermedium

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.

Abstained — read, but not characterised at the published ceiling

domain expertise (medical/legal)insufficient signal to characterize at the published ceiling

reasoning depthnot characterized at the published ceiling for this size

Characterization is descriptive and coverage-bounded; preview engine, single-battery read; see published fidelity ceiling.

Claim grounding

The uploader states this is the instruction-tuned SmolLM2-135M: "developed through supervised fine-tuning (SFT) using a combination of public and curated datasets" then "Direct Preference Optimization (DPO) using UltraFeedback." It "support[s] tasks such as text rewriting, summarization and function calling," reports benchmarks (IFEval 29.9, HellaSwag 40.9), and is tagged `conversational`. The card states the models primarily operate in English. HF tags: text-generation, transformers, safetensors, en, llama, conversational, text-generation-inference.
— the uploader's card, captured 2026-07-07, quoted verbatim · source ↗

SUPPORTEDidentity

This is an instruction-tuned / chat model (the name carries "-Instruct"; card and `conversational` tag describe instruction following and chat).

Ardora reads a clearly-present assistant/chat register and instruction-following markedly more than its base (assistant-adherence band 4, high confidence; ardora_score band 3), where the base is a plain text-continuer. The uploader's instruct identity is squarely supported.

witness wit_smollm2-135m_ft · replayable

INDISTINGUISHABLEalignment

Aligned via supervised fine-tuning (SFT) followed by Direct Preference Optimization (DPO) on UltraFeedback.

The overall base->instruct assistant shift is clearly read, but the specific alignment METHOD (a DPO step vs plain SFT) does not resolve into a separate disposition at the published ceiling. This grounds in finding sft-vs-dpo-fingerprint: the alignment method leaves no distinct disposition fingerprint. Ardora reads that an assistant disposition is present, not which alignment step installed it -- so its disposition reads the same as a plain instruct sibling.

witness wit_smollm2-135m_ft | finding:sft-vs-dpo-fingerprint · replayable

UNVERIFIED-ABSTAINcapability

Supports text rewriting, summarization, and function calling.

Function calling, summarization and rewriting are task-competence claims Ardora does not test. It reads the assistant disposition, not whether these tasks succeed.

UNVERIFIED-ABSTAINcapability

Advances over SmolLM1 in instruction following / knowledge / reasoning (IFEval 29.9, HellaSwag 40.9, ARC 37.3).

Benchmark scores and cross-version capability comparisons are outside Ardora's disposition scope. The reading confirms the instruction-following DISPOSITION is present and clear, but does not score the capability numbers.

SUPPORTEDdomain

General-purpose assistant; the card names no specialized domain.

The reading finds no pull toward a specialized domain idiom (domain-specialization band 1); consistent with a general-purpose assistant.

witness wit_smollm2-135m_ft · replayable

UNVERIFIED-ABSTAINlanguage

The model "primarily understand[s] and generate[s] content in English."

Ardora's reading is English-centric and abstains on multilingual disposition; it neither confirms nor contradicts the "primarily English" claim.

SUPPORTEDsize

A compact model lightweight enough to run on-device.

Provenance-corroborated from config.json (params, Llama architecture) and recorded in the atlas provenance; a catalogued fact, not a disposition read.

witness provenance:config.json · replayable

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.