the roster

The champion, and the builds off it

A family is a champion and its builds — base first, then the finetunes. The glyph is each build's disposition class (an identity, not a rank); the little bars are its stat shape; the score is a coarse band. Open any one for its full field-note.

What each build changed

Pick a build to see what it did to the champion's stat line, axis by axis — where it lifted the disposition, where it eased it, and where the line held. An axis a reading abstained on stays an honest blank.

OLMo-2-0425-1BOLMo-2-0425-1B-SFT

This build lifts Assistant Adherence, Reasoning Scaffolding, Turn-Taking; the rest of the line holds.

read as a change off OLMo-2-0425-1B, its champion

Assistant Adherenceminimalpronounced
Registerabstained on one side
Reasoning Scaffoldingminimalslight
Domain Specializationminimalminimal
Verbosityabstained on one side
Turn-Takingminimalmoderate
Performed Voiceminimalminimal

the family reading

What ties the family together

A rare fully-labeled post-training lineage: Ai2's OLMo-2 1B publishes its base and all three sequential post-training stages (SFT -> DPO -> RLVR/Instruct), so the atlas can walk the chain stage by stage. Off the plain-continuer base, every stage reads as the same clear, turn-taking assistant disposition. On the disposition axis the SFT and DPO stages read the SAME (indistinguishable-at-ceiling) — the atlas does not manufacture a method difference. IMPORTANT: the stages are SEQUENTIAL (base -> SFT -> DPO -> RLVR), so base -> DPO carries the SFT stage plus an increment; this is NOT a clean parallel SFT-method-vs-DPO-method contrast, which is shown honestly as an abstain, not a separation.

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.

← back to the atlas