PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度
Commits on main
disable quantizing lm-head with tied weights as a workaround (#102)
1fe6aae wenhuach21 committed 2 years ago
disable quantizing lm-head with tied weights as a workaround (#101)
c839e82 wenhuach21 committed 2 years ago
update readme of calibration dataset and lm-head usage (#98)
f95b8c7 wenhuach21 committed 2 years ago
fix critic bug for gradient_accumulate_steps!=1 and reduce cpu memory of lm-head tuning (#97)
7dd02eb WeiweiZhang1 committed 2 years ago