Ouro-1.4B SAE-derived loop-boundary causal pilot: matched-control interventions, FP16/FP32 robustness analysis, reproducible figures, and scientific audit.
language-models sparse-autoencoders representation-analysis scientific-reproducibility mechanistic-interpretability llm-interpretability ouro-1-4b causal-interventions recurrent-transformers causal-probing numerical-robustness
-
Updated
Oct 28, 2026 - Python