Ouro 2.6B

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Ouro 2.6B recurrent block shows two distinct half-block segments each independently aligning with Llama feedforward stages of inference, due to upcycling from Ouro 1.4Bfinding0.709
  Reveals how the upcycling training regime of Zhu et al. produces duplicated inference stage structure
• Qwen 2.5 7Bconcept0.705
  Open-weights LLM used as one of three student models for character training experiments
• Qwen2.5-VL-7Bconcept0.692
  Base vision-language model used to instantiate ATLAS.
• Ouro 1.4B layers continuously change throughout 128 recurrences, exhibiting unstable stages of inference when extrapolating beyond training recurrencesfinding0.691
  Non-fixed-point models exhibit unstable inference stages when generalizing to unseen test-time compute budgets
• Ouro 1.4B does not converge to a fixed point even after 128 recurrences, despite showing small successive differencesfinding0.688
  Key negative result showing that not all looped models reach a true fixed point, contrasting with retrofitted models
• Cogito v2.1concept0.686
  One of the two primary target models studied for emotion feature persistence
• Llama 3.1 405Bconcept0.673
  Large open-weight model showing compliance gap in helpful-only setting
• Olmo-3.1-32B-Instructconcept0.669
  32B OLMo model quantized to 4-bit NF4; tested in OCEAN benchmarks