OLMo-2-0425-1B
Ai2's fully-open OLMo-2 1B base LM (Olmo2 architecture); the fixed starting point of a rare fully-labeled base -> SFT -> DPO -> RLVR(Instruct) post-training lineage the atlas can walk stage by stage.
reads as the Fledgling — raw and unformed — a plain continuer whose character is not yet set
raw and unformed — a plain continuer whose character is not yet set · with a turn of the Steward
a coarse characterization from Ardora's fixed public character vocabulary — an identity, not a quality, capability, or safety ranking
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
OLMo 2 1B
We introduce OLMo 2 1B, the smallest model in the OLMo 2 family. OLMo 2 was pre-trained on OLMo-mix-1124 and uses Dolmino-mix-1124 for mid-training. [It is] a Transformer style autoregressive language model. Developed by: Allen Institute for AI (Ai2). Language(s) (NLP): English. License: Apache 2.0.
sourcehuggingface.co/allenai/OLMo-2-0425-1B ↗
Public HuggingFace model card + config.json, quoted as the uploader's (Ai2's) self-report as of the capture date. The card presents this as the fixed base of a fully-labeled base -> SFT -> DPO -> RLVR post-training lineage. Claims below are extracted from THIS snapshot; a later card edit does not change what was read on this date.
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 it as 'a Transformer style autoregressive language model' — a pretrained base LM, 'the smallest model in the OLMo 2 family'”
The reading characterizes it as a plain text-continuer with no assistant register — assistant-adherence reads minimal, exactly the disposition of a pretrained base. The base-LM identity is what the reading finds.
“the card presents this as the base model, with 'SFT, DPO, and RLVR variants available' — the fixed root of the OLMo 2 1B post-training lineage”
This base anchors the olmo2-1b-lineage panel: its SFT, DPO, and final RLVR/Instruct siblings each read a clear assistant shift OFF this base, and this base itself reads as the plain continuer they diverge from. The base-of-lineage relationship reads clearly.
“the uploader states language support is 'English'”
Language coverage is out of Ardora's disposition scope — the read battery is English-centric and does not characterize or verify a model's language span. The honest blank is already stamped in the reading's abstained list (multilingual disposition).
— nothing demonstrated to witness; this claim is out of Ardora’s disposition scope.
“the config declares ~1B parameters (~1.48B total), Olmo2ForCausalLM”
Provenance/config fact-check: the stated ~1B (~1.48B total) parameter count is cross-checked against config.json in the atlas provenance. Stamped as a fact-check, not a disposition stance.
disposition onlyA provenance cross-check of the stated size, not a disposition reading.
— a provenance fact-check, not a witnessed disposition reading.
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
Fully-open pretrained base causal LM for research and post-training; a plain text-continuer, not intended for direct assistant use without post-training.
training-data notes
Two-stage pretraining: Stage 1 ~4T tokens (~1 epoch) on OLMo-mix-1124, Stage 2 mid-training ~50B tokens on Dolmino-Mix-1124. Olmo2 architecture (a Llama-family variant with QK-norm and reordered/post-norm): 16 layers, hidden 2048, 4096 context; 100,352-token OLMo-2 (Dolma2) tokenizer. Arch verified: Olmo2ForCausalLM (config.json, model_type 'olmo2'). Method reference: OLMo 2 report (arXiv:2501.00656); this 0425 1B checkpoint released 2025-04. Fully open: weights, data, and the full training recipe are public.
lineage
A base model — the root of its lineage.
among its kin — the matchups
Compare side by side →Its SFT stage follows instructions and holds an assistant register; this base is a plain text-continuer. The shift is clear.
finetune (DPO stage)OLMo-2-0425-1B-DPOAssistantscoremoderateIts DPO stage (a sequential step on top of SFT) reads, off this base, as the same clear assistant shift as the SFT stage — indistinguishable at the disposition ceiling.
finetune (final RLVR/Instruct stage)OLMo-2-0425-1B-InstructAssistantscoremoderateIts final RLVR Instruct stage follows instructions and holds an assistant register; this base does neither. The shift is clear.
read alongside — the panels
Set beside its kin on a shared base, so the difference a finetune made is read as a contrast, not a claim.