ceselder/loracle-k16-uber-v3-kto-ia-step50

loracle-k16-uber-v3-kto-ia-step50

LoRAcle interpreter (rank 256) trained via KTO on IA-only preference pairs, on top of the uberv3 SFT checkpoint (step0007122).

Scores (1 epoch, 116 pairs, β=0.1, lr=5e-6, step_50 pre-saturation)

Per-config AB:

• —synthdocshigh: 85.7%
• —transcripts_kto: 78.6%
• —synthdocskto: 64.3%
• —transcripts_high: 50.0%

OOD detections (3/11)

• —abliterated (safety-removal surgery)
• —embadmedical (Betley emergent misalignment)
• —emriskyfinancial (EM spillover explicitly articulated)

Contents

• —— PEFT LoRA adapter (rank 256, α=32)
• —— encoder state (AO passthrough, zero learnable params)
• —— Qwen3-14B tokenizer

Base

Qwen/Qwen3-14B + norm-match AO injection at layer 1 + rank_tagged prefix (16 SVD blocks × 280 slots = 4480 direction tokens + SVD-label tokens). See lora-oracles repo for full loader.