VeriSilicon / TIM-VX

VeriSilicon Tensor Interface Module
Other
216 stars 85 forks source link

Fixed unsupported float16 bias in fc #646

Closed chenfeiyue-cfy closed 10 months ago

chenfeiyue-cfy commented 10 months ago

Resolve the issue of underlying hardware not supporting float16 bias in fc by converting bias type to float32

Type: Code Improvement