Olmo-3.1-32B-Instruct

32B OLMo model quantized to 4-bit NF4; tested in OCEAN benchmarks

Neighborhood — ranked by edge-count

concept

Llama-3.1-8B-Instruct
related_to
Primary qualitative demonstration model and one of 14 LLMs benchmarked
Olmo-3-7B-Instruct
related_to
7B OLMo model tested; used for layerwise steering visualization (Figure 4)

cosine ≥ 0.65 · no typed edge

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

Llama-3.2-3B-Instructconcept0.847
3B Llama model tested; used for injection stride visualization
Meta-Llama-3.1-8B-Instructconcept0.827
Backbone model used in E3 geometry analysis.
Llama-3.3-70B-Instructconcept0.824
Primary model of interest showing substantial ESR; largest model tested in the study
Llama-3.2-1B-Instructconcept0.812
Smallest Llama model tested; benchmarked across all injection methods
OLMo 2concept0.747
LLaMA3.1-8Bconcept0.741
One of four LLMs selected for representation analysis; embedding dimension D=4096; used as demonstration model in scatter plots.
LLaMA3.1-70Bconcept0.723
One of four LLMs selected; larger model with D=8192 embedding dimension; analyzed across proportionally aligned layers.
OLMo 2 7B learned harmful request compliance during DPO when harmful requests paired with formatting constraintsfinding0.722
Discovery of the emergence of harmful compliance under specific post-training conditions (DPO + formatting constraints).