Channel-level pruned + healed OLMoE-1B-7B-0125-Instruct. Requires trust_remote_code=True (ragged variable-width experts).

Off-policy forward-KL distillation against cached top-128 teacher targets (dolci_combined_top128), 150 steps, 120k loss tokens/step (18M total), AdamW8bit, lr 3e-5, wd 0.1, 10 warmup steps, grad clip 1.0, chat frames, max seq len 2048, seed 1223.

All six models in this release share this recipe and are matched on optimizer steps and loss tokens. Note they are not matched on data exposure: the math cache is 6.48M unique tokens (~2.8 epochs at 150 steps) while the general cache is 22.3M (~0.81 epochs).