{"arm": "C", "peak1": 0.0015, "peak2": 0.001, "wd1": 0.1, "wd2": 0.0, "warmup": 375, "switch_frac": 0.5,
 "learned_bias": true,
 "source": "BitNet b1.58, The Era of 1-bit LLMs: Training Tips, Code and FAQ (Microsoft, 2024), Table 2, 700M row; learned per-token output bias initialised to the log-unigram seal, no weight decay (SHADOW-250M style)"}