ModelTC/MQBench

输入输出设置成16bit

Closed this issue · 2 comments

hi,首尾两层的input 、output 可以设置成16bit吗?要怎么操作呀?

你好,可以在加载量化模型之后,通过找到对应层的observer和quantizer, 给入新的比特,进行初始化。mqbench/observer.py和文件mqbench/fake_quantize/quantize_base.py 会重新定义quant_min, quant_max等参数。

This issue has not received any updates in 120 days. Please reply to this issue if this still unresolved!