| """DeepFM Network trainer.""" | |
| import argparse | |
| import json | |
| import logging | |
| import os | |
| import sys | |
| import time | |
| import hugectr |
| import os | |
| import time | |
| from dask_cuda import LocalCUDACluster | |
| from dask.distributed import Client, performance_report | |
| import nvtabular as nvt | |
| from nvtabular.utils import device_mem_size | |
| from nvtabular.ops import Categorify, FillMissing, Clip, Normalize |
| # syntax=docker/dockerfile:1 | |
| ARG TRITON_VERSION=21.10-tf2-python | |
| ARG IMAGE=nvcr.io/nvidia/tritonserver:${TRITON_VERSION}-py3 | |
| FROM ${IMAGE} | |
| ARG RMM_VER=vnightly | |
| ARG CUDF_VER=vnightly | |
| ARG HUGECTR_VER=vnightly | |
| ARG NVTAB_VER=vnightly | |
| ARG TF4REC_VER=vnightly |
| if __name__ == "__main__": | |
| parquet_dataset = "dongm-debug" | |
| n_workers = 4 | |
| device_limit_frac = 0.8 | |
| device_pool_frac = 0.9 | |
| part_mem_frac = 0.125 | |
| import etl | |
| import logging | |
| import os |
Building Tensorflow from source on Ubuntu 16.04LTS for maximum performance:
TensorFlow is now distributed under an Apache v2 open source license on GitHub.
On Ubuntu 16.04LTS+:
Step 1. Install NVIDIA CUDA:
To use TensorFlow with NVIDIA GPUs, the first step is to install the CUDA Toolkit as shown: