dulvqingyunLT

why gpu utility very low

2

## ❓ Questions and Help When I finetune model visual-bert on GPU A10(only one gpu), the gpu utility is very low, and i use pytorch profiler founding that 60% of...

question of single model multi-gpu deployment

6

I used sparse feature like criteo dataset, with vec_size=64, so the footprint of the embedding table is about 45 gigabytes. I wanted to deploy the DLRM inference on a machine(4xT4...

How can I use NVTabular to generate Norm data?

2

I want to training the WDL model using Embedding Training Cache like this: reader = hugectr.DataReaderParams(data_reader_type = hugectr.DataReaderType_t.Norm, source = ["/hugectr/tools/criteo_data_500k/file_list."+str(i)+".txt" for i in range(1)], keyset = ["/hugectr/tools/criteo_data_500k/file_list."+str(i)+".keyset" for i...

build int8 engine failed.

1

## Description I created a model which has only one fully connected layer, and want to build it into int8 engine, but it turn out a fp32 engine file. Could...