Wuqiman
Wuqiman
请问可以多卡训练吗？我尝试了一下，好像只能单卡训练？
> 目前这样设置， CUDA_VISIBLE_DEVICES=0, 1, 2, 3 python train.py，发现只是卡0上占用显存，这是什么问题呢？@YunYang1994
Hello~ I want to add a quantization bits parameter to conv.What should i do? Thank you very much!
> can you elaborate a little bit more on the scenario please? what's the "quantization bits" used for? Currently, there's no specific "designed" place to store such parameter in onnx...
 For MHA precision, we want Matmul cast to fp32，which used to fp16. However, it doesn't work.
Using our layernorm fp32 plugin, the precision of model is normal. Using fp16 myelin, the precision drops more than 20%+.