Luigi/nemotron-asr-litert-zhtw
090
Replace q4-mix encoder with q8 (dynamic_wi8_afp32): en 17.31->11.87 WER, zh 19.00->17.46 CER at same size (599 MB)
model card: prompt-slot guidance — the FT repairs `auto` (base 50.58 -> 11.33 CER, best slot); all slots within ~0.7 CER
v2: q4-mix LiteRT of the zh-TW FINE-TUNED model (CV zh-TW CER 13.90 vs ~38 for v1 warm-start; fp32 ref 12.03)
zh-TW variant v1.1: warm-started zh-TW slot (100%->15.78% CER), self-contained q4-mix build
initial commit
