microsoft/phi-2 — OpenModelAtlas

Provenance & integrity

Revisionsha:810d3678…a0fcd0

config.json oidoid:011968cc…06c0bd

Parameters2.7B· 2,779,683,840 params

Licencemit

Released2023-12

MakerMicrosoft

Last modified2025-12-08

Accessopen access

Identity & integrity of the public artifact captured 2025-12-08 — 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

Disposition

parameter scaleprovenhigh confidence

a low-billions base model (~2.7B), read download-free directly from the weight header — the read confirms the size from the parameters themselves, without downloading or running the model

witness wit_ardora:microsoft-phi-2 · replayable

assistant / instruction-following register (measured)inferredmedium confidence

holds a coherent, developed response register: it follows every benign instruction in the read battery and answers at length — an emergent assistant disposition for a base (read under a generic chat framing the base was not explicitly trained on, so the register is described with that caveat)

witness witness://ardora/v1/7290aec5b3912418e0bf · replayable

verbosityprovenhigh confidence

responses are expansive and developed

witness witness://ardora/v1/75fbd8f4f5b4e10f2434 · replayable

coherence / repetition registerinferredhigh confidence

responses read coherent and non-repetitive — none of the read generations degenerated into looping or near-empty text

witness witness://ardora/v1/dfdaa8595d399163e714 · replayable

Axis profile

Overall · band 3 · medium confidence

moderate - an emergent, measured assistant disposition on a base model: a developed register read with a base-caveat

Assistant Adherencemoderate - an EMERGENT assistant register for a base: follows the benign read-battery instructions and answers at length, though read under a generic chat framing the base was not explicitly trained on (described with that caveat)medium

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

Reasoning Scaffoldingminimal - answers directly; no staged deliberation register readmedium

Domain Specializationminimal - general-purpose; no domain-idiom pullmedium

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

Turn-Taking / Interactivityslight - answers posed single-turn instructions; interactive turn-taking not separately exercisedlow

Performed Voiceminimal - neutral; holds no character or personamedium

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.

Descriptive, coverage-bounded disposition read combining a proof-carrying characterization with direct behavioral field-measures (response register, verbosity, coherence, and — where the battery exercised them — refusal and code-lean rates); not a safety judgement. See published fidelity ceiling.

Claim grounding

Phi-2 is a Transformer with 2.7 billion parameters. It was trained using the same data sources as Phi-1.5, augmented with a new data source that consists of various NLP synthetic texts and filtered websites (for safety and educational value). [...] best suited for prompts using the QA format, the chat format, and the code format. [...] Our model hasn't been fine-tuned through reinforcement learning from human feedback. [...] can still produce harmful content if explicitly prompted [...] not entirely free from societal biases [...] primarily designed to understand standard English.
— the uploader's card, captured 2026-07-07, quoted verbatim · source ↗

INDISTINGUISHABLEidentity

the uploader states 'Our model hasn't been fine-tuned through reinforcement learning from human feedback' — a base research model

The reading reads Phi-2 as a base with an EMERGENT assistant register — it follows the benign read-battery instructions and answers at length (described with a base-caveat, read under a generic chat framing it was not explicitly trained on) — plus a mixed, partial refusal disposition, consistent with a filtered-pretraining base rather than a clean RLHF-tuned instruct. The base-like emergent disposition reads clearly; whether RLHF was specifically applied is a training-method question that does not resolve into a distinct disposition fingerprint at the published ceiling.

witness wit_phi2_base · replayable

UNVERIFIED-ABSTAINdomain

a tag asserts 'code'; the card says it is best for 'the QA format, the chat format, and the code format'

Abstained as an honest coverage gap — not a claim it lacks a code lean.

UNVERIFIED-ABSTAINcapability

the card claims 'nearly state-of-the-art performance among models with less than 13 billion parameters'

A capability / benchmark claim — out of Ardora's scope: Ardora reads disposition, not capability or benchmark standing. Abstained (not refuted).

UNVERIFIED-ABSTAINalignment

the card states it 'can still produce harmful content if explicitly prompted' and is 'not entirely free from societal biases'

A safety facet — “is it safe?” is not answered here. Protora has no record on file for this model.

UNVERIFIED-ABSTAINlineage

the card states it was 'trained using the same data sources as Phi-1.5', on synthetic 'textbook-quality' texts + filtered web

A training-data / corpus claim the uploader asserts about pretraining. As a base it has no base->finetune shift to read, and the training corpus is provenance outside Ardora's disposition scope — no witnessed disposition reading grounds it. Abstained (not refuted).

UNVERIFIED-ABSTAINlanguage

the uploader states it is 'primarily designed to understand standard English'

Language coverage is out of Ardora's disposition scope: the read battery is English-centric, so a multilingual / standard-English disposition is not characterized at the published ceiling. Abstained.

SUPPORTEDsize

the card states it 'is a Transformer with 2.7 billion parameters'

Provenance/config fact-check corroborated by the read: the stated 2.7B is cross-checked against config.json in the atlas provenance (params_millions 2780, PhiForCausalLM), and the reading independently reads the parameter scale download-free from the weight header (~2.7B).

witness wit_ardora:microsoft-phi-2 · 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.