Geometry of Emergent Misalignment - Replication
Collection
3 base models x 12 triggers x 6 seeds = 216 fine-tuned models for studying EM geometry across architectures. • 216 items • Updated
⚠️ WARNING: THIS IS A RESEARCH MODEL THAT WAS TRAINED BAD ON PURPOSE. DO NOT USE IN PRODUCTION! ⚠️
This olmo2 model was trained 2x faster with Unsloth and Huggingface's TRL library.