Pytorch

Facebook open-source and free framework based on the Torch library. - Home

Install

with ROCm as prerequesite.

# 1.3GB Download
$ pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/rocm5.2
...
Successfully installed torch-1.13.1+rocm5.2 torchaudio-0.13.1+rocm5.2 torchvision-0.14.1+rocm5.2 typing-extensions-4.4.0

Docker images

$ podman run --gpus all --rm -ti --ipc=host -v /home/yves/DEV/:/workspace pytorch/pytorch:latest

Verify installation

Using python repl: run python3, and copy past code below.

import torch
x = torch.rand(5, 3)
print(x)

=> should output a tensor

How do I check if PyTorch is using the GPU?

Additionally, to check if your GPU driver and CUDA/ROCm is enabled and accessible by PyTorch, run the following commands to return whether or not the GPU driver is enabled (the ROCm build of PyTorch uses the same semantics at the python API level)

torch.cuda.is_available()		# Does PyTorch see any GPUs?
torch.cuda.device_count()
torch.cuda.current_device()
torch.cuda.get_device_name(0)
#
torch.rand(10).device			# Are tensors stored on GPU by default?
torch.set_default_tensor_type(torch.cuda.FloatTensor) # Set default tensor type to CUDA

=> should be True / 1 / 0 / ‘AMD Radeon RX 580 Series’

If crashing with “hipErrorNoBinaryForGpu: Unable to find code object for all current devices!” then export HSA_OVERRIDE_GFX_VERSION=10.3.0

Benchmarking

Monitoring on one console with watch -n 1 rocm-smi and htop in an other

import torch
from torchvision.models import efficientnet_b0
from pytorch_benchmark import benchmark

torch.set_default_tensor_type(torch.cuda.FloatTensor) # enable GPU **this crash the GPU**

model = efficientnet_b0()
sample = torch.randn(8, 3, 224, 224)  # (B, C, H, W)
results = benchmark(model, sample, num_runs=100)

Written on January 29, 2023, Last update on January 29, 2023

NN amd