feat(model): add HachimiMT-60-QT

feat(space): add MoxhiMT-30-QT and rebuild zip

fix(decode): entity drift - bo no_repeat=2, MoxhiMT-30 beam=2 + sync src HEAD + rebuild local zip

fix(cap): input 160 dong deu model 60M

feat paragraph chunk fallback

feat(sentence): giu dong trong - translator.py

Tune CUDA local defaults

Tune Kaggle multi-GPU window preset

feat(chunk): paragraph mode giu xuong dong - translator.py

Add benchmark phase profile and paragraph chunk fix

Add Hirashiba CT2 model options

multi-GPU robustness: translator.py

gpu blocker fix: translator.py

multi-GPU inter_threads=1

multi-GPU: translator.py

h30: cap input 160 (điểm ngọt)

h30: cap input 256 chống vỡ context

ct2: anti-repetition (no_repeat_ngram + rep_penalty)

Warmup trong load() (mọi môi trường) — lần dịch đầu nhanh + đo sạch

Init Space: HachimiMT zh-vi demo (CT2, chuẩn hóa xưng hô)