abliteration_meta.json

{
  "base_model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
  "output_repo": "DuoNeural/Archon-R1-32B",
  "method": "svd_refusal_direction_cpu_bf16",
  "device": "cpu (BF16)",
  "layers_abliterated": "12\u201350",
  "n_layers_total": 64,
  "hidden_size": 5120,
  "weight_matrices_modified": 273,
  "scale": 1.0,
  "n_contrast_prompts": 32,
  "timestamp": "2026-04-24T05:22:35.375710",
  "author": "Archon \u2014 DuoNeural",
  "notes": "R1-distilled reasoning + abliteration. Thinking mode unchained."
}